Summary of Tasks and Exercises Completed for Final Deliverables¶
Ex 1.1: Load the real dataset images and preparing the data for training and Visualizing Sample Images from the Training Dataset with Labels¶
Ex 2.1: Create and Train a Custom Neural Network from Scratch¶
Ex 2.2: Build, Train, and Evaluate ANN and CNN Models Using PyTorch¶
Ex 3.1: Building and Evaluating ANN and CNN Models with TensorFlow¶
Ex 4.1: Transfer Learning approach with ResNet50: Training the Model [resnet50+additional layers] from Scratch.¶
Ex 4.2: Transfer Learning with ResNet50: Model Training, Evaluation, and Results Using TensorFlow¶
Ex 4.3: Comparing Optimizers: Evaluating Test Accuracy for Various Optimization Algorithms.¶
Ex 5.1: Transfer Learning with VGG16: Training, Evaluating, and Saving the Model¶
Ex 6.1: Training and Testing Multiple Pretrained Models for this Image Classification problem¶
Ex 7.1: Using Active learning and verifying its effect on models performance.¶
Exercise 1.1: Load the real dataset images and preparing the data for training and Visualizing Sample Images from the Training Dataset with Labels¶
Unrefined data¶
In [5]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras.preprocessing.image import ImageDataGenerator
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 32
datagen = ImageDataGenerator(rescale=1./255)
train_generator = datagen.flow_from_directory(
'Unrefined\\Train',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
validation_generator = datagen.flow_from_directory(
'Unrefined\\Validation',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_generator = datagen.flow_from_directory(
'Unrefined\\Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
def plot_images(images_arr, labels):
fig, axes = plt.subplots(1, len(images_arr), figsize=(40,40))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title('Without Mask' if label == 1 else 'With Mask')
plt.tight_layout()
plt.show()
images, labels = next(train_generator)
plot_images(images[:10], labels[:10])
Found 8022 images belonging to 2 classes. Found 1746 images belonging to 2 classes. Found 2024 images belonging to 2 classes.
In [6]:
import seaborn as sns
import pandas as pd
def plot_combined_class_distribution(train_generator, validation_generator, test_generator):
def get_class_distribution(generator, name):
labels = generator.classes
class_names = list(generator.class_indices.keys())
label_counts = np.bincount(labels)
return pd.DataFrame({
'Class': class_names,
'Count': label_counts,
'Dataset': name
})
train_dist = get_class_distribution(train_generator, 'Train')
validation_dist = get_class_distribution(validation_generator, 'Validation')
test_dist = get_class_distribution(test_generator, 'Test')
combined_dist = pd.concat([train_dist, validation_dist, test_dist])
plt.figure(figsize=(12, 8))
sns.barplot(x='Class', y='Count', hue='Dataset', data=combined_dist, palette='viridis')
plt.title('Class Distribution across Unrefined Datasets')
plt.ylabel('Number of Images')
plt.xlabel('Class')
plt.legend(title='Dataset')
plt.show()
plot_combined_class_distribution(train_generator, validation_generator, test_generator)
Refined data¶
In [1]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras.preprocessing.image import ImageDataGenerator
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 32
datagen = ImageDataGenerator(rescale=1./255)
train_generator = datagen.flow_from_directory(
'Train',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
validation_generator = datagen.flow_from_directory(
'Validation',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_generator = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
def plot_images(images_arr, labels):
fig, axes = plt.subplots(1, len(images_arr), figsize=(40,40))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title('Without Mask' if label == 1 else 'With Mask')
plt.tight_layout()
plt.show()
images, labels = next(train_generator)
plot_images(images[:10], labels[:10])
Found 2999 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes.
Ex 1.2: Visualizing Class Distribution Across Training, Validation, and Test Datasets¶
In [4]:
import seaborn as sns
import pandas as pd
def plot_combined_class_distribution(train_generator, validation_generator, test_generator):
def get_class_distribution(generator, name):
labels = generator.classes
class_names = list(generator.class_indices.keys())
label_counts = np.bincount(labels)
return pd.DataFrame({
'Class': class_names,
'Count': label_counts,
'Dataset': name
})
train_dist = get_class_distribution(train_generator, 'Train')
validation_dist = get_class_distribution(validation_generator, 'Validation')
test_dist = get_class_distribution(test_generator, 'Test')
combined_dist = pd.concat([train_dist, validation_dist, test_dist])
plt.figure(figsize=(12, 8))
sns.barplot(x='Class', y='Count', hue='Dataset', data=combined_dist, palette='viridis')
plt.title('Class Distribution across Refined Datasets')
plt.ylabel('Number of Images')
plt.xlabel('Class')
plt.legend(title='Dataset')
plt.show()
plot_combined_class_distribution(train_generator, validation_generator, test_generator)
In [ ]:
import numpy as np
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
train_dir = 'Train'
val_dir = 'Validation'
test_dir = 'Test'
train_gen = ImageDataGenerator(rescale=1.0/255.0)
val_gen = ImageDataGenerator(rescale=1.0/255.0)
test_gen = ImageDataGenerator(rescale=1.0/255.0)
train_data = train_gen.flow_from_directory(
train_dir,
target_size=(32, 32),
batch_size=32,
class_mode='binary',
shuffle=False
)
val_data = val_gen.flow_from_directory(
val_dir,
target_size=(32, 32),
batch_size=32,
class_mode='binary',
shuffle=False
)
test_data = test_gen.flow_from_directory(
test_dir,
target_size=(32, 32),
batch_size=32,
class_mode='binary',
shuffle=False
)
def get_data(generator):
imgs = []
lbls = []
for batch in generator:
img_batch, lbl_batch = batch
imgs.append(img_batch)
lbls.append(lbl_batch)
if len(imgs) * generator.batch_size >= generator.samples:
break
return np.concatenate(imgs), np.concatenate(lbls)
train_imgs, train_lbls = get_data(train_data)
val_imgs, val_lbls = get_data(val_data)
test_imgs, test_lbls = get_data(test_data)
train_imgs_flat = train_imgs.reshape(train_imgs.shape[0], -1)
val_imgs_flat = val_imgs.reshape(val_imgs.shape[0], -1)
test_imgs_flat = test_imgs.reshape(test_imgs.shape[0], -1)
input_size = 32 * 32 * 3
hidden1_size = 128
hidden2_size = 64
output_size = 1
np.random.seed(42)
w1 = np.random.randn(input_size, hidden1_size) * np.sqrt(2. / input_size)
b1 = np.zeros((1, hidden1_size))
w2 = np.random.randn(hidden1_size, hidden2_size) * np.sqrt(2. / hidden1_size)
b2 = np.zeros((1, hidden2_size))
w3 = np.random.randn(hidden2_size, output_size) * np.sqrt(2. / hidden2_size)
b3 = np.zeros((1, output_size))
def relu(x):
return np.maximum(0, x)
def relu_derivative(x):
return np.where(x > 0, 1, 0)
def sigmoid(x):
return 1 / (1 + np.exp(-x))
def sigmoid_derivative(x):
s = sigmoid(x)
return s * (1 - s)
def binary_cross_entropy(y_true, y_pred):
epsilon = 1e-15
y_pred = np.clip(y_pred, epsilon, 1 - epsilon)
loss = - np.mean(y_true * np.log(y_pred) + (1 - y_true) * np.log(1 - y_pred))
return loss
def binary_cross_entropy_derivative(y_true, y_pred):
epsilon = 1e-15
y_pred = np.clip(y_pred, epsilon, 1 - epsilon)
grad = y_pred - y_true.reshape(-1, 1)
return grad / y_true.shape[0]
def forward_pass(X):
z1 = np.dot(X, w1) + b1
a1 = relu(z1)
z2 = np.dot(a1, w2) + b2
a2 = relu(z2)
z3 = np.dot(a2, w3) + b3
a3 = sigmoid(z3)
return z1, a1, z2, a2, z3, a3
def backward_pass(X, y, z1, a1, z2, a2, z3, a3, lr=0.01):
global w1, b1, w2, b2, w3, b3
dz3 = binary_cross_entropy_derivative(y, a3)
dw3 = np.dot(a2.T, dz3)
db3 = np.sum(dz3, axis=0, keepdims=True)
dz2 = np.dot(dz3, w3.T) * relu_derivative(z2)
dw2 = np.dot(a1.T, dz2)
db2 = np.sum(dz2, axis=0, keepdims=True)
dz1 = np.dot(dz2, w2.T) * relu_derivative(z1)
dw1 = np.dot(X.T, dz1)
db1 = np.sum(dz1, axis=0, keepdims=True)
w1 -= lr * dw1
b1 -= lr * db1
w2 -= lr * dw2
b2 -= lr * db2
w3 -= lr * dw3
b3 -= lr * db3
epochs = 30
learning_rate = 0.01
train_acc_list = []
val_acc_list = []
val_split = 0.1
num_val_samples = int(val_split * train_imgs_flat.shape[0])
val_imgs_split = train_imgs_flat[:num_val_samples]
val_lbls_split = train_lbls[:num_val_samples]
train_imgs_split = train_imgs_flat[num_val_samples:]
train_lbls_split = train_lbls[num_val_samples:]
for epoch in range(epochs):
z1, a1, z2, a2, z3, a3 = forward_pass(train_imgs_split)
loss = binary_cross_entropy(train_lbls_split, a3)
backward_pass(train_imgs_split, train_lbls_split, z1, a1, z2, a2, z3, a3, learning_rate)
train_preds = (a3 > 0.5).astype(int).flatten()
train_acc = np.mean(train_preds == train_lbls_split)
train_acc_list.append(train_acc)
z1_val, a1_val, z2_val, a2_val, z3_val, a3_val = forward_pass(val_imgs_split)
val_preds = (a3_val > 0.5).astype(int).flatten()
val_acc = np.mean(val_preds == val_lbls_split)
val_acc_list.append(val_acc)
print(f'Epoch {epoch+1}/{epochs}, Train Accuracy: {train_acc:.4f}, Val Accuracy: {val_acc:.4f}')
_, _, _, _, _, a3_test = forward_pass(test_imgs_flat)
test_preds = (a3_test > 0.5).astype(int).flatten()
test_acc = np.mean(test_preds == test_lbls)
print(f'Test accuracy of custom built model: {test_acc:.4f}')
Found 8191 images belonging to 2 classes. Found 1817 images belonging to 2 classes. Found 1784 images belonging to 2 classes. Epoch 1/30, Train Accuracy: 0.5619, Val Accuracy: 0.2979 Epoch 2/30, Train Accuracy: 0.5883, Val Accuracy: 0.2344 Epoch 3/30, Train Accuracy: 0.6135, Val Accuracy: 0.3150 Epoch 4/30, Train Accuracy: 0.6466, Val Accuracy: 0.3211 Epoch 5/30, Train Accuracy: 0.6645, Val Accuracy: 0.3736 Epoch 6/30, Train Accuracy: 0.6894, Val Accuracy: 0.4017 Epoch 7/30, Train Accuracy: 0.7075, Val Accuracy: 0.4396 Epoch 8/30, Train Accuracy: 0.7294, Val Accuracy: 0.4701 Epoch 9/30, Train Accuracy: 0.7432, Val Accuracy: 0.4982 Epoch 10/30, Train Accuracy: 0.7615, Val Accuracy: 0.5263 Epoch 11/30, Train Accuracy: 0.7729, Val Accuracy: 0.5482 Epoch 12/30, Train Accuracy: 0.7855, Val Accuracy: 0.5702 Epoch 13/30, Train Accuracy: 0.7976, Val Accuracy: 0.5897 Epoch 14/30, Train Accuracy: 0.8078, Val Accuracy: 0.6032 Epoch 15/30, Train Accuracy: 0.8151, Val Accuracy: 0.6203 Epoch 16/30, Train Accuracy: 0.8216, Val Accuracy: 0.6361 Epoch 17/30, Train Accuracy: 0.8313, Val Accuracy: 0.6484 Epoch 18/30, Train Accuracy: 0.8353, Val Accuracy: 0.6532 Epoch 19/30, Train Accuracy: 0.8405, Val Accuracy: 0.6593 Epoch 20/30, Train Accuracy: 0.8443, Val Accuracy: 0.6740 Epoch 21/30, Train Accuracy: 0.8490, Val Accuracy: 0.6825 Epoch 22/30, Train Accuracy: 0.8531, Val Accuracy: 0.6838 Epoch 23/30, Train Accuracy: 0.8555, Val Accuracy: 0.6923 Epoch 24/30, Train Accuracy: 0.8570, Val Accuracy: 0.6996 Epoch 25/30, Train Accuracy: 0.8587, Val Accuracy: 0.7009 Epoch 26/30, Train Accuracy: 0.8607, Val Accuracy: 0.7045 Epoch 27/30, Train Accuracy: 0.8641, Val Accuracy: 0.7094 Epoch 28/30, Train Accuracy: 0.8653, Val Accuracy: 0.7143 Epoch 29/30, Train Accuracy: 0.8676, Val Accuracy: 0.7192 Epoch 30/30, Train Accuracy: 0.8698, Val Accuracy: 0.7253 Test accuracy: 0.8660
In [ ]:
plt.figure(figsize=(10, 6))
plt.plot(range(1, epochs + 1), train_accuracy_list, label='Training Accuracy')
plt.plot(range(1, epochs + 1), val_accuracy_list, label='Validation Accuracy')
plt.axhline(y=test_accuracy, color='r', linestyle='--', label='Test Accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.title('Training, Validation, and Test Accuracy over 30 Epochs for custom built model.')
plt.legend()
plt.tight_layout()
plt.show()
Ex 2.2: Build, Train, and Evaluate ANN and CNN Models Using PyTorch¶
In this experiment, I built, trained, and evaluated two types of neural networks—an artificial neural network (ANN) and a convolutional neural network (CNN)—using PyTorch. This was done out of curiosity to explore and experiment with PyTorch for the first time.¶
In [6]:
import torch
import torch.nn.functional as F
from torch.utils.data import DataLoader, Dataset
from torchvision import datasets, transforms
import numpy as np
import torch.nn as nn
class MyDataset(Dataset):
def __init__(self, directory, transform=None):
self.data = datasets.ImageFolder(directory, transform=transform)
def __len__(self):
return len(self.data)
def __getitem__(self, index):
img, label = self.data[index]
img = np.array(img).astype(np.float32).transpose((2, 0, 1)) / 255.0
return torch.tensor(img), torch.tensor(label, dtype=torch.long)
transform = transforms.Compose([
transforms.Resize((128, 128)),
])
train_data = MyDataset('Train', transform=transform)
val_data = MyDataset('Validation', transform=transform)
test_data = MyDataset('Test', transform=transform)
train_loader = DataLoader(train_data, batch_size=32, shuffle=True)
val_loader = DataLoader(val_data, batch_size=32, shuffle=False)
test_loader = DataLoader(test_data, batch_size=32, shuffle=False)
class SimpleANN(nn.Module):
def __init__(self):
super(SimpleANN, self).__init__()
self.fc1 = nn.Linear(128*128*3, 256)
self.fc2 = nn.Linear(256, 128)
self.fc3 = nn.Linear(128, 64)
self.fc4 = nn.Linear(64, 32)
self.fc5 = nn.Linear(32, 1)
def forward(self, x):
x = x.view(x.size(0), -1)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.relu(self.fc3(x))
x = F.relu(self.fc4(x))
x = torch.sigmoid(self.fc5(x))
return x
class SimpleCNN(nn.Module):
def __init__(self):
super(SimpleCNN, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=3, padding=1)
self.conv2 = nn.Conv2d(64, 128, kernel_size=3, padding=1)
self.conv3 = nn.Conv2d(128, 256, kernel_size=3, padding=1)
self.conv4 = nn.Conv2d(256, 512, kernel_size=3, padding=1)
self.fc1 = nn.Linear(512 * 8 * 8, 256)
self.fc2 = nn.Linear(256, 128)
self.fc3 = nn.Linear(128, 64)
self.fc4 = nn.Linear(64, 32)
self.fc5 = nn.Linear(32, 1)
def forward(self, x):
x = F.relu(self.conv1(x))
x = F.max_pool2d(x, 2)
x = F.relu(self.conv2(x))
x = F.max_pool2d(x, 2)
x = F.relu(self.conv3(x))
x = F.max_pool2d(x, 2)
x = F.relu(self.conv4(x))
x = F.max_pool2d(x, 2)
x = x.view(x.size(0), -1)
x = F.relu(self.fc1(x))
x = F.relu(self.fc2(x))
x = F.relu(self.fc3(x))
x = F.relu(self.fc4(x))
x = torch.sigmoid(self.fc5(x))
return x
def train(model, loader, epochs=10, learning_rate=0.001):
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
criterion = torch.nn.BCELoss()
for epoch in range(epochs):
model.train()
total_loss = 0.0
correct = 0
total = 0
for images, labels in loader:
optimizer.zero_grad()
outputs = model(images)
labels = labels.float().unsqueeze(1)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
total_loss += loss.item()
preds = outputs > 0.5
correct += (preds.float() == labels).sum().item()
total += labels.size(0)
accuracy = correct / total
print(f'Epoch [{epoch+1}/{epochs}], Loss: {total_loss/len(loader)}, Accuracy: {accuracy}')
def evaluate(model, loader):
model.eval()
correct = 0
total = 0
with torch.no_grad():
for images, labels in loader:
outputs = model(images)
preds = outputs > 0.5
correct += (preds.float() == labels.unsqueeze(1).float()).sum().item()
total += labels.size(0)
accuracy = correct / total
print(f'Test Accuracy: {accuracy}')
ann_model = SimpleANN()
cnn_model = SimpleCNN()
print("Training ANN Model:")
train(ann_model, train_loader)
print("Training CNN Model:")
train(cnn_model, train_loader)
print("Evaluating ANN Model:")
evaluate(ann_model, test_loader)
print("Evaluating CNN Model:")
evaluate(cnn_model, test_loader)
Training ANN Model: Epoch [1/10], Loss: 0.4177964595705271, Accuracy: 0.80375 Epoch [2/10], Loss: 0.2478776265308261, Accuracy: 0.90125 Epoch [3/10], Loss: 0.21687977977097034, Accuracy: 0.9146875 Epoch [4/10], Loss: 0.1985765428096056, Accuracy: 0.92375 Epoch [5/10], Loss: 0.15627139957621694, Accuracy: 0.94375 Epoch [6/10], Loss: 0.1575373009033501, Accuracy: 0.94125 Epoch [7/10], Loss: 0.160260842256248, Accuracy: 0.934375 Epoch [8/10], Loss: 0.12530433816835285, Accuracy: 0.9525 Epoch [9/10], Loss: 0.1347218832047656, Accuracy: 0.9515625 Epoch [10/10], Loss: 0.1100270338030532, Accuracy: 0.9553125 Training CNN Model: Epoch [1/10], Loss: 0.3670965425297618, Accuracy: 0.80125 Epoch [2/10], Loss: 0.14400036776438355, Accuracy: 0.9478125 Epoch [3/10], Loss: 0.09360392600065097, Accuracy: 0.970625 Epoch [4/10], Loss: 0.08267202332615853, Accuracy: 0.971875 Epoch [5/10], Loss: 0.05977901781719993, Accuracy: 0.979375 Epoch [6/10], Loss: 0.05942668042029254, Accuracy: 0.980625 Epoch [7/10], Loss: 0.06066181360045448, Accuracy: 0.980625 Epoch [8/10], Loss: 0.05993445452535525, Accuracy: 0.980625 Epoch [9/10], Loss: 0.042700811686299855, Accuracy: 0.986875 Epoch [10/10], Loss: 0.02464769098442048, Accuracy: 0.9921875 Evaluating ANN Model: Test Accuracy: 0.8703125 Evaluating CNN Model: Test Accuracy: 0.9609375
Now I used above trained PyTorch model to load and preprocess new images, make predictions, and visualize the results. The goal was to test the model's performance on unseen data and categorize images as "With Mask" or "Without Mask."¶
In [8]:
import os
import numpy as np
import torch
from PIL import Image
import matplotlib.pyplot as plt
from torchvision import transforms
def preprocess_image(img_path):
img = Image.open(img_path).convert('RGB')
transform = transforms.Compose([
transforms.Resize((128, 128)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
img_tensor = transform(img)
img_tensor = img_tensor.unsqueeze(0)
return img_tensor
image_dir = 'NewTest'
image_files = [os.path.join(image_dir, f) for f in os.listdir(image_dir)]
mask_images = []
no_mask_images = []
model = cnn_model
model.eval()
for img_path in image_files:
try:
img_tensor = preprocess_image(img_path)
with torch.no_grad():
prediction = model(img_tensor).item()
label = 'Without Mask' if prediction > 0.5 else 'With Mask'
print(f"Image: {img_path}, Prediction: {prediction:.4f}, Label: {label}")
img = Image.open(img_path).resize((128, 128))
if label == 'Without Mask':
no_mask_images.append(np.array(img))
else:
mask_images.append(np.array(img))
except Exception as e:
print(f"Error with image {img_path}: {e}")
def display_images(img_arr, labels):
fig, ax = plt.subplots(1, len(img_arr), figsize=(20, 20))
ax = ax.flatten()
for img, a, label in zip(img_arr, ax, labels):
a.imshow(img)
a.axis('off')
a.set_title(label)
plt.tight_layout()
plt.show()
if mask_images:
display_images(mask_images, ['With Mask'] * len(mask_images))
if no_mask_images:
display_images(no_mask_images, ['Without Mask'] * len(no_mask_images))
Image: NewTest\178.png, Prediction: 0.0000, Label: With Mask Image: NewTest\187.png, Prediction: 0.9918, Label: Without Mask Image: NewTest\190.png, Prediction: 0.0000, Label: With Mask Image: NewTest\219.png, Prediction: 0.9631, Label: Without Mask Image: NewTest\231.png, Prediction: 0.9995, Label: Without Mask Image: NewTest\283.png, Prediction: 1.0000, Label: Without Mask Image: NewTest\350.png, Prediction: 0.0000, Label: With Mask Image: NewTest\351.png, Prediction: 0.0000, Label: With Mask Image: NewTest\69.png, Prediction: 1.0000, Label: Without Mask Image: NewTest\72.png, Prediction: 1.0000, Label: Without Mask Image: NewTest\download (1).jpg, Prediction: 0.0576, Label: With Mask Image: NewTest\download (2).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\download (3).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\download (4).jpg, Prediction: 0.0002, Label: With Mask Image: NewTest\download (6).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\download (7).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\download.jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\face-13.jpg, Prediction: 0.9997, Label: Without Mask Image: NewTest\images (2).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\images (3).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\images (4).jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\images.jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\jh.jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\kjk.jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\lk.jpg, Prediction: 0.0000, Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file 'NewTest\\Thumbs.db' Image: NewTest\uu.jpg, Prediction: 0.0000, Label: With Mask Image: NewTest\with.jpg, Prediction: 0.0000, Label: With Mask
In [5]:
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Flatten, Conv2D, MaxPooling2D, Dropout
import math
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 32
datagen = ImageDataGenerator(rescale=1./255)
train_generator = datagen.flow_from_directory(
'Train',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
validation_generator = datagen.flow_from_directory(
'Validation',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_generator = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
ann_model = Sequential([
Flatten(input_shape=(IMG_HEIGHT, IMG_WIDTH, 3)),
Dense(256, activation='relu'),
Dense(128, activation='relu'),
Dense(64, activation='relu'),
Dense(32, activation='relu'),
Dense(1, activation='sigmoid')
])
ann_model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
cnn_model = Sequential([
Conv2D(64, (3, 3), activation='relu', input_shape=(IMG_HEIGHT, IMG_WIDTH, 3)),
MaxPooling2D(pool_size=(2, 2)),
Conv2D(128, (3, 3), activation='relu'),
MaxPooling2D(pool_size=(2, 2)),
Conv2D(256, (3, 3), activation='relu'),
MaxPooling2D(pool_size=(2, 2)),
Conv2D(512, (3, 3), activation='relu'),
MaxPooling2D(pool_size=(2, 2)),
Flatten(),
Dense(256, activation='relu'),
Dropout(0.5),
Dense(128, activation='relu'),
Dense(64, activation='relu'),
Dense(32, activation='relu'),
Dense(1, activation='sigmoid')
])
cnn_model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
ann_history = ann_model.fit(
train_generator,
epochs=10,
validation_data=validation_generator,
)
cnn_history = cnn_model.fit(
train_generator,
epochs=10,
validation_data=validation_generator,
)
ann_loss, ann_accuracy = ann_model.evaluate(test_generator)
print(f'ANN Model - Loss: {ann_loss}, Accuracy: {ann_accuracy}')
cnn_loss, cnn_accuracy = cnn_model.evaluate(test_generator)
print(f'CNN Model - Loss: {cnn_loss}, Accuracy: {cnn_accuracy}')
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes.
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\layers\reshaping\flatten.py:37: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead. super().__init__(**kwargs) C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\layers\convolutional\base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead. super().__init__(activity_regularizer=activity_regularizer, **kwargs)
Epoch 1/10
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
100/100 ━━━━━━━━━━━━━━━━━━━━ 69s 640ms/step - accuracy: 0.7047 - loss: 1.3946 - val_accuracy: 0.8219 - val_loss: 0.4435 Epoch 2/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 65s 605ms/step - accuracy: 0.8778 - loss: 0.3364 - val_accuracy: 0.8641 - val_loss: 0.3715 Epoch 3/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 67s 628ms/step - accuracy: 0.8793 - loss: 0.3339 - val_accuracy: 0.8672 - val_loss: 0.3576 Epoch 4/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 69s 658ms/step - accuracy: 0.9113 - loss: 0.2351 - val_accuracy: 0.8734 - val_loss: 0.3998 Epoch 5/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 68s 656ms/step - accuracy: 0.9114 - loss: 0.2441 - val_accuracy: 0.8797 - val_loss: 0.3348 Epoch 6/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 69s 656ms/step - accuracy: 0.9334 - loss: 0.1692 - val_accuracy: 0.8844 - val_loss: 0.3357 Epoch 7/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 68s 642ms/step - accuracy: 0.9352 - loss: 0.1672 - val_accuracy: 0.8844 - val_loss: 0.3419 Epoch 8/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 70s 657ms/step - accuracy: 0.9355 - loss: 0.1756 - val_accuracy: 0.9016 - val_loss: 0.2472 Epoch 9/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 64s 598ms/step - accuracy: 0.9426 - loss: 0.1590 - val_accuracy: 0.8859 - val_loss: 0.3639 Epoch 10/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 68s 637ms/step - accuracy: 0.8976 - loss: 0.2801 - val_accuracy: 0.8547 - val_loss: 0.4111 Epoch 1/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 77s 708ms/step - accuracy: 0.7445 - loss: 0.4675 - val_accuracy: 0.9469 - val_loss: 0.2384 Epoch 2/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 76s 719ms/step - accuracy: 0.9597 - loss: 0.1377 - val_accuracy: 0.9500 - val_loss: 0.1403 Epoch 3/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 78s 746ms/step - accuracy: 0.9522 - loss: 0.1374 - val_accuracy: 0.9438 - val_loss: 0.2017 Epoch 4/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 80s 767ms/step - accuracy: 0.9695 - loss: 0.1227 - val_accuracy: 0.9672 - val_loss: 0.0994 Epoch 5/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 73s 684ms/step - accuracy: 0.9782 - loss: 0.0870 - val_accuracy: 0.9719 - val_loss: 0.0898 Epoch 6/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 71s 681ms/step - accuracy: 0.9729 - loss: 0.0759 - val_accuracy: 0.9656 - val_loss: 0.0997 Epoch 7/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 73s 683ms/step - accuracy: 0.9859 - loss: 0.0453 - val_accuracy: 0.9812 - val_loss: 0.0516 Epoch 8/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 74s 697ms/step - accuracy: 0.9910 - loss: 0.0364 - val_accuracy: 0.9719 - val_loss: 0.1164 Epoch 9/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 73s 685ms/step - accuracy: 0.9835 - loss: 0.0509 - val_accuracy: 0.9563 - val_loss: 0.1394 Epoch 10/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 70s 660ms/step - accuracy: 0.9860 - loss: 0.0474 - val_accuracy: 0.9656 - val_loss: 0.1968 20/20 ━━━━━━━━━━━━━━━━━━━━ 9s 480ms/step - accuracy: 0.8785 - loss: 0.3243 ANN Model - Loss: 0.34392955899238586, Accuracy: 0.8734375238418579 20/20 ━━━━━━━━━━━━━━━━━━━━ 4s 193ms/step - accuracy: 0.9686 - loss: 0.1222 CNN Model - Loss: 0.12716740369796753, Accuracy: 0.96875
In [6]:
import matplotlib.pyplot as plt
plt.figure(figsize=(12, 6))
plt.plot(ann_history.history['accuracy'], label='ANN Training Accuracy')
plt.plot(ann_history.history['val_accuracy'], label='ANN Validation Accuracy')
plt.plot(cnn_history.history['accuracy'], label='CNN Training Accuracy')
plt.plot(cnn_history.history['val_accuracy'], label='CNN Validation Accuracy')
plt.ylim([0.5,1.0])
plt.axhline(y=ann_accuracy, color='r', linestyle='--', label='ANN Test Accuracy')
plt.axhline(y=cnn_accuracy, color='b', linestyle='--', label='CNN Test Accuracy')
plt.title('Model Accuracy of ANN and CNN')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.show()
In [14]:
test_gen = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary',
shuffle=False
)
from sklearn.metrics import confusion_matrix, classification_report
import seaborn as sns
import matplotlib.pyplot as plt
cnn_loss, cnn_acc = cnn_model.evaluate(test_gen, verbose=2)
print(f'CNN Model - Loss: {cnn_loss}, Accuracy: {cnn_acc}')
test_gen.reset()
y_pred = cnn_model.predict(test_gen, verbose=2)
y_pred_classes = (y_pred > 0.5).astype(int).flatten()
y_true = test_gen.classes
print(f'y_pred: {y_pred[:10]}')
print(f'y_pred_classes: {y_pred_classes[:10]}')
print(f'y_true: {y_true[:10]}')
cm = confusion_matrix(y_true, y_pred_classes)
cm_labels = list(test_gen.class_indices.keys())
plt.figure(figsize=(8, 6))
sns.heatmap(cm, annot=True, fmt='d', xticklabels=cm_labels, yticklabels=cm_labels, cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title('Confusion Matrix')
plt.show()
print(classification_report(y_true, y_pred_classes, target_names=cm_labels))
Found 640 images belonging to 2 classes.
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
20/20 - 10s - 482ms/step - accuracy: 0.9688 - loss: 0.1272 CNN Model - Loss: 0.12716737389564514, Accuracy: 0.96875 20/20 - 4s - 196ms/step Y_pred: [[4.9713572e-07] [1.1781770e-09] [1.1037769e-07] [1.0646973e-03] [1.7775730e-03] [5.2717024e-07] [2.3479188e-02] [5.1347720e-16] [7.0326034e-09] [1.6866721e-01]] Y_pred_classes: [0 0 0 0 0 0 0 0 0 0] Y_true: [0 0 0 0 0 0 0 0 0 0]
precision recall f1-score support
WithMask 0.98 0.96 0.97 320
WithoutMask 0.96 0.98 0.97 320
accuracy 0.97 640
macro avg 0.97 0.97 0.97 640
weighted avg 0.97 0.97 0.97 640
In [12]:
plt.figure(figsize=(12, 6))
plt.subplot(1, 2, 1)
plt.plot(ann_history.history['loss'], label='ANN Training Loss')
plt.plot(ann_history.history['val_loss'], label='ANN Validation Loss')
plt.plot(cnn_history.history['loss'], label='CNN Training Loss')
plt.plot(cnn_history.history['val_loss'], label='CNN Validation Loss')
plt.title('Model Loss Over Epochs')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.ylim([0,1.0])
plt.subplot(1, 2, 2)
plt.plot(ann_history.history['accuracy'], label='ANN Training Accuracy')
plt.plot(ann_history.history['val_accuracy'], label='ANN Validation Accuracy')
plt.plot(cnn_history.history['accuracy'], label='CNN Training Accuracy')
plt.plot(cnn_history.history['val_accuracy'], label='CNN Validation Accuracy')
plt.title('Model Accuracy Over Epochs')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend()
plt.ylim([0,1.0])
plt.show()
Predicting New Images with the Trained CNN Model¶
In [17]:
def preprocess_image(img_path):
img = tf.keras.utils.load_img(img_path, target_size=(IMG_HEIGHT, IMG_WIDTH))
img_array = tf.keras.utils.img_to_array(img)
img_array = tf.expand_dims(img_array, axis=0)
img_array /= 255.0
return img_array
def predict_and_display_image(img_path, cnn_model, threshold=0.5):
img_array = preprocess_image(img_path)
prediction = cnn_model.predict(img_array)
plt.imshow(tf.keras.utils.load_img(img_path))
plt.axis('off')
if prediction[0] > threshold:
plt.title(f'Predicted: Without Mask ({prediction[0][0]:.4f})')
else:
plt.title(f'Predicted: With Mask ({1 - prediction[0][0]:.4f})')
plt.show()
new_image_path = 'NewTest\with.jpg'
predict_and_display_image(new_image_path, cnn_model, threshold=0.6)
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 138ms/step
In [ ]:
new_image_path = 'NewTest\283.png'
predict_and_display_image(new_image_path, cnn_model, threshold=0.6)
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 30ms/step
In [ ]:
new_image_path = 'NewTest/th.jpg'
predict_and_display_image(new_image_path, cnn_model, threshold=0.6)
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 33ms/step
In [20]:
new_image_path = 'NewTest/face-13.jpg'
predict_and_display_image(new_image_path, cnn_model, threshold=0.6)
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 36ms/step
Ex 4.1: Transfer Learning approach with ResNet50: Training the Model [resnet50+additional layers] from Scratch.¶
In [ ]:
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import ResNet50
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
from math import ceil
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 128
EPOCHS = 10
LEARNING_RATE = 0.001
datagen = ImageDataGenerator(rescale=1./255)
train_gen = datagen.flow_from_directory(
'Train',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
val_gen = datagen.flow_from_directory(
'Validation',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_gen = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
train_samples = len(train_gen.filenames)
val_samples = len(val_gen.filenames)
steps_per_epoch = ceil(train_samples / BATCH_SIZE)
validation_steps = ceil(val_samples / BATCH_SIZE)
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(IMG_HEIGHT, IMG_WIDTH, 3))
base_model.trainable = False
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(128, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(64, activation='relu')(x)
pred = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=pred)
optimizer = tf.keras.optimizers.Adam(learning_rate=LEARNING_RATE)
loss_fn = tf.keras.losses.BinaryCrossentropy()
def train_step(images, labels):
with tf.GradientTape() as tape:
preds = model(images, training=True)
loss = loss_fn(labels, preds)
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
accuracy = tf.keras.metrics.binary_accuracy(labels, preds)
return loss, accuracy
def train_model(train_gen, val_gen, epochs):
history = {'loss': [], 'accuracy': [], 'val_loss': [], 'val_accuracy': []}
for epoch in range(epochs):
print(f'Epoch {epoch + 1}/{epochs}')
epoch_loss = 0
epoch_accuracy = 0
num_samples = 0
for step in range(steps_per_epoch):
batch_images, batch_labels = next(train_gen)
batch_loss, batch_accuracy = train_step(batch_images, batch_labels)
batch_size = batch_images.shape[0]
epoch_loss += batch_loss.numpy() * batch_size
epoch_accuracy += tf.reduce_sum(batch_accuracy).numpy()
num_samples += batch_size
print(f'Step {step + 1} of Epoch {epoch + 1} completed')
epoch_loss /= num_samples
epoch_accuracy /= num_samples
history['loss'].append(epoch_loss)
history['accuracy'].append(epoch_accuracy)
val_loss, val_accuracy = 0, 0
num_samples = 0
for step in range(validation_steps):
batch_images, batch_labels = next(val_gen)
preds = model(batch_images, training=False)
loss = loss_fn(batch_labels, preds)
batch_size = batch_images.shape[0]
val_loss += loss.numpy() * batch_size
val_accuracy += tf.reduce_sum(tf.keras.metrics.binary_accuracy(batch_labels, preds)).numpy()
num_samples += batch_size
print(f'Step {step + 1} of Validation in Epoch {epoch + 1} completed')
val_loss /= num_samples
val_accuracy /= num_samples
history['val_loss'].append(val_loss)
history['val_accuracy'].append(val_accuracy)
print(f'Loss: {epoch_loss:.4f}, Accuracy: {epoch_accuracy:.4f}, Val Loss: {val_loss:.4f}, Val Accuracy: {val_accuracy:.4f}')
return history
history = train_model(train_gen, val_gen, EPOCHS)
def evaluate_model(test_gen):
test_loss, test_accuracy = 0, 0
num_samples = 0
for step in range(steps_per_epoch):
batch_images, batch_labels = next(test_gen)
preds = model(batch_images, training=False)
loss = loss_fn(batch_labels, preds)
batch_size = batch_images.shape[0]
test_loss += loss.numpy() * batch_size
test_accuracy += tf.reduce_sum(tf.keras.metrics.binary_accuracy(batch_labels, preds)).numpy()
num_samples += batch_size
print(f'Step {step + 1} of Testing completed')
test_loss /= num_samples
test_accuracy /= num_samples
return test_loss, test_accuracy
test_loss, test_accuracy = evaluate_model(test_gen)
print(f'Test accuracy: {test_accuracy:.4f}')
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes. Epoch 1/10 Step 1 of Epoch 1 completed Step 2 of Epoch 1 completed Step 3 of Epoch 1 completed Step 4 of Epoch 1 completed Step 5 of Epoch 1 completed Step 6 of Epoch 1 completed Step 7 of Epoch 1 completed Step 8 of Epoch 1 completed Step 9 of Epoch 1 completed Step 10 of Epoch 1 completed Step 11 of Epoch 1 completed Step 12 of Epoch 1 completed Step 13 of Epoch 1 completed Step 14 of Epoch 1 completed Step 15 of Epoch 1 completed Step 16 of Epoch 1 completed Step 17 of Epoch 1 completed Step 18 of Epoch 1 completed Step 19 of Epoch 1 completed Step 20 of Epoch 1 completed Step 21 of Epoch 1 completed Step 22 of Epoch 1 completed Step 23 of Epoch 1 completed Step 24 of Epoch 1 completed Step 25 of Epoch 1 completed Step 1 of Validation in Epoch 1 completed Step 2 of Validation in Epoch 1 completed Step 3 of Validation in Epoch 1 completed Step 4 of Validation in Epoch 1 completed Step 5 of Validation in Epoch 1 completed Loss: 0.6924, Accuracy: 0.5134, Val Loss: 0.6866, Val Accuracy: 0.5000 Epoch 2/10 Step 1 of Epoch 2 completed Step 2 of Epoch 2 completed Step 3 of Epoch 2 completed Step 4 of Epoch 2 completed Step 5 of Epoch 2 completed Step 6 of Epoch 2 completed Step 7 of Epoch 2 completed Step 8 of Epoch 2 completed Step 9 of Epoch 2 completed Step 10 of Epoch 2 completed Step 11 of Epoch 2 completed Step 12 of Epoch 2 completed Step 13 of Epoch 2 completed Step 14 of Epoch 2 completed Step 15 of Epoch 2 completed Step 16 of Epoch 2 completed Step 17 of Epoch 2 completed Step 18 of Epoch 2 completed Step 19 of Epoch 2 completed Step 20 of Epoch 2 completed Step 21 of Epoch 2 completed Step 22 of Epoch 2 completed Step 23 of Epoch 2 completed Step 24 of Epoch 2 completed Step 25 of Epoch 2 completed Step 1 of Validation in Epoch 2 completed Step 2 of Validation in Epoch 2 completed Step 3 of Validation in Epoch 2 completed Step 4 of Validation in Epoch 2 completed Step 5 of Validation in Epoch 2 completed Loss: 0.6655, Accuracy: 0.6222, Val Loss: 0.6583, Val Accuracy: 0.5938 Epoch 3/10 Step 1 of Epoch 3 completed Step 2 of Epoch 3 completed Step 3 of Epoch 3 completed Step 4 of Epoch 3 completed Step 5 of Epoch 3 completed Step 6 of Epoch 3 completed Step 7 of Epoch 3 completed Step 8 of Epoch 3 completed Step 9 of Epoch 3 completed Step 10 of Epoch 3 completed Step 11 of Epoch 3 completed Step 12 of Epoch 3 completed Step 13 of Epoch 3 completed Step 14 of Epoch 3 completed Step 15 of Epoch 3 completed Step 16 of Epoch 3 completed Step 17 of Epoch 3 completed Step 18 of Epoch 3 completed Step 19 of Epoch 3 completed Step 20 of Epoch 3 completed Step 21 of Epoch 3 completed Step 22 of Epoch 3 completed Step 23 of Epoch 3 completed Step 24 of Epoch 3 completed Step 25 of Epoch 3 completed Step 1 of Validation in Epoch 3 completed Step 2 of Validation in Epoch 3 completed Step 3 of Validation in Epoch 3 completed Step 4 of Validation in Epoch 3 completed Step 5 of Validation in Epoch 3 completed Loss: 0.6242, Accuracy: 0.6744, Val Loss: 0.7006, Val Accuracy: 0.5062 Epoch 4/10 Step 1 of Epoch 4 completed Step 2 of Epoch 4 completed Step 3 of Epoch 4 completed Step 4 of Epoch 4 completed Step 5 of Epoch 4 completed Step 6 of Epoch 4 completed Step 7 of Epoch 4 completed Step 8 of Epoch 4 completed Step 9 of Epoch 4 completed Step 10 of Epoch 4 completed Step 11 of Epoch 4 completed Step 12 of Epoch 4 completed Step 13 of Epoch 4 completed Step 14 of Epoch 4 completed Step 15 of Epoch 4 completed Step 16 of Epoch 4 completed Step 17 of Epoch 4 completed Step 18 of Epoch 4 completed Step 19 of Epoch 4 completed Step 20 of Epoch 4 completed Step 21 of Epoch 4 completed Step 22 of Epoch 4 completed Step 23 of Epoch 4 completed Step 24 of Epoch 4 completed Step 25 of Epoch 4 completed Step 1 of Validation in Epoch 4 completed Step 2 of Validation in Epoch 4 completed Step 3 of Validation in Epoch 4 completed Step 4 of Validation in Epoch 4 completed Step 5 of Validation in Epoch 4 completed Loss: 0.5897, Accuracy: 0.6963, Val Loss: 0.5903, Val Accuracy: 0.6625 Epoch 5/10 Step 1 of Epoch 5 completed Step 2 of Epoch 5 completed Step 3 of Epoch 5 completed Step 4 of Epoch 5 completed Step 5 of Epoch 5 completed Step 6 of Epoch 5 completed Step 7 of Epoch 5 completed Step 8 of Epoch 5 completed Step 9 of Epoch 5 completed Step 10 of Epoch 5 completed Step 11 of Epoch 5 completed Step 12 of Epoch 5 completed Step 13 of Epoch 5 completed Step 14 of Epoch 5 completed Step 15 of Epoch 5 completed Step 16 of Epoch 5 completed Step 17 of Epoch 5 completed Step 18 of Epoch 5 completed Step 19 of Epoch 5 completed Step 20 of Epoch 5 completed Step 21 of Epoch 5 completed Step 22 of Epoch 5 completed Step 23 of Epoch 5 completed Step 24 of Epoch 5 completed Step 25 of Epoch 5 completed Step 1 of Validation in Epoch 5 completed Step 2 of Validation in Epoch 5 completed Step 3 of Validation in Epoch 5 completed Step 4 of Validation in Epoch 5 completed Step 5 of Validation in Epoch 5 completed Loss: 0.5479, Accuracy: 0.7256, Val Loss: 0.5289, Val Accuracy: 0.7688 Epoch 6/10 Step 1 of Epoch 6 completed Step 2 of Epoch 6 completed Step 3 of Epoch 6 completed Step 4 of Epoch 6 completed Step 5 of Epoch 6 completed Step 6 of Epoch 6 completed Step 7 of Epoch 6 completed Step 8 of Epoch 6 completed Step 9 of Epoch 6 completed Step 10 of Epoch 6 completed Step 11 of Epoch 6 completed Step 12 of Epoch 6 completed Step 13 of Epoch 6 completed Step 14 of Epoch 6 completed Step 15 of Epoch 6 completed Step 16 of Epoch 6 completed Step 17 of Epoch 6 completed Step 18 of Epoch 6 completed Step 19 of Epoch 6 completed Step 20 of Epoch 6 completed Step 21 of Epoch 6 completed Step 22 of Epoch 6 completed Step 23 of Epoch 6 completed Step 24 of Epoch 6 completed Step 25 of Epoch 6 completed Step 1 of Validation in Epoch 6 completed Step 2 of Validation in Epoch 6 completed Step 3 of Validation in Epoch 6 completed Step 4 of Validation in Epoch 6 completed Step 5 of Validation in Epoch 6 completed Loss: 0.5257, Accuracy: 0.7334, Val Loss: 0.4962, Val Accuracy: 0.7656 Epoch 7/10 Step 1 of Epoch 7 completed Step 2 of Epoch 7 completed Step 3 of Epoch 7 completed Step 4 of Epoch 7 completed Step 5 of Epoch 7 completed Step 6 of Epoch 7 completed Step 7 of Epoch 7 completed Step 8 of Epoch 7 completed Step 9 of Epoch 7 completed Step 10 of Epoch 7 completed Step 11 of Epoch 7 completed Step 12 of Epoch 7 completed Step 13 of Epoch 7 completed Step 14 of Epoch 7 completed Step 15 of Epoch 7 completed Step 16 of Epoch 7 completed Step 17 of Epoch 7 completed Step 18 of Epoch 7 completed Step 19 of Epoch 7 completed Step 20 of Epoch 7 completed Step 21 of Epoch 7 completed Step 22 of Epoch 7 completed Step 23 of Epoch 7 completed Step 24 of Epoch 7 completed Step 25 of Epoch 7 completed Step 1 of Validation in Epoch 7 completed Step 2 of Validation in Epoch 7 completed Step 3 of Validation in Epoch 7 completed Step 4 of Validation in Epoch 7 completed Step 5 of Validation in Epoch 7 completed Loss: 0.4755, Accuracy: 0.7875, Val Loss: 0.5599, Val Accuracy: 0.6813 Epoch 8/10 Step 1 of Epoch 8 completed Step 2 of Epoch 8 completed Step 3 of Epoch 8 completed Step 4 of Epoch 8 completed Step 5 of Epoch 8 completed Step 6 of Epoch 8 completed Step 7 of Epoch 8 completed Step 8 of Epoch 8 completed Step 9 of Epoch 8 completed Step 10 of Epoch 8 completed Step 11 of Epoch 8 completed Step 12 of Epoch 8 completed Step 13 of Epoch 8 completed Step 14 of Epoch 8 completed Step 15 of Epoch 8 completed Step 16 of Epoch 8 completed Step 17 of Epoch 8 completed Step 18 of Epoch 8 completed Step 19 of Epoch 8 completed Step 20 of Epoch 8 completed Step 21 of Epoch 8 completed Step 22 of Epoch 8 completed Step 23 of Epoch 8 completed Step 24 of Epoch 8 completed Step 25 of Epoch 8 completed Step 1 of Validation in Epoch 8 completed Step 2 of Validation in Epoch 8 completed Step 3 of Validation in Epoch 8 completed Step 4 of Validation in Epoch 8 completed Step 5 of Validation in Epoch 8 completed Loss: 0.4359, Accuracy: 0.8087, Val Loss: 0.4356, Val Accuracy: 0.8328 Epoch 9/10 Step 1 of Epoch 9 completed Step 2 of Epoch 9 completed Step 3 of Epoch 9 completed Step 4 of Epoch 9 completed Step 5 of Epoch 9 completed Step 6 of Epoch 9 completed Step 7 of Epoch 9 completed Step 8 of Epoch 9 completed Step 9 of Epoch 9 completed Step 10 of Epoch 9 completed Step 11 of Epoch 9 completed Step 12 of Epoch 9 completed Step 13 of Epoch 9 completed Step 14 of Epoch 9 completed Step 15 of Epoch 9 completed Step 16 of Epoch 9 completed Step 17 of Epoch 9 completed Step 18 of Epoch 9 completed Step 19 of Epoch 9 completed Step 20 of Epoch 9 completed Step 21 of Epoch 9 completed Step 22 of Epoch 9 completed Step 23 of Epoch 9 completed Step 24 of Epoch 9 completed Step 25 of Epoch 9 completed Step 1 of Validation in Epoch 9 completed Step 2 of Validation in Epoch 9 completed Step 3 of Validation in Epoch 9 completed Step 4 of Validation in Epoch 9 completed Step 5 of Validation in Epoch 9 completed Loss: 0.4108, Accuracy: 0.8250, Val Loss: 0.4174, Val Accuracy: 0.8094 Epoch 10/10 Step 1 of Epoch 10 completed Step 2 of Epoch 10 completed Step 3 of Epoch 10 completed Step 4 of Epoch 10 completed Step 5 of Epoch 10 completed Step 6 of Epoch 10 completed Step 7 of Epoch 10 completed Step 8 of Epoch 10 completed Step 9 of Epoch 10 completed Step 10 of Epoch 10 completed Step 11 of Epoch 10 completed Step 12 of Epoch 10 completed Step 13 of Epoch 10 completed Step 14 of Epoch 10 completed Step 15 of Epoch 10 completed Step 16 of Epoch 10 completed Step 17 of Epoch 10 completed Step 18 of Epoch 10 completed Step 19 of Epoch 10 completed Step 20 of Epoch 10 completed Step 21 of Epoch 10 completed Step 22 of Epoch 10 completed Step 23 of Epoch 10 completed Step 24 of Epoch 10 completed Step 25 of Epoch 10 completed Step 1 of Validation in Epoch 10 completed Step 2 of Validation in Epoch 10 completed Step 3 of Validation in Epoch 10 completed Step 4 of Validation in Epoch 10 completed Step 5 of Validation in Epoch 10 completed Loss: 0.3924, Accuracy: 0.8275, Val Loss: 0.4560, Val Accuracy: 0.7984 Step 1 of Testing completed Step 2 of Testing completed Step 3 of Testing completed Step 4 of Testing completed Step 5 of Testing completed Step 6 of Testing completed Step 7 of Testing completed Step 8 of Testing completed Step 9 of Testing completed Step 10 of Testing completed Step 11 of Testing completed Step 12 of Testing completed Step 13 of Testing completed Step 14 of Testing completed Step 15 of Testing completed Step 16 of Testing completed Step 17 of Testing completed Step 18 of Testing completed Step 19 of Testing completed Step 20 of Testing completed Step 21 of Testing completed Step 22 of Testing completed Step 23 of Testing completed Step 24 of Testing completed Step 25 of Testing completed Test accuracy: 0.8156
In [ ]:
epochs_range = range(1, EPOCHS + 1)
plt.figure(figsize=(12, 6))
plt.plot(epochs_range, history['accuracy'], label='Training Accuracy')
plt.plot(epochs_range, history['val_accuracy'], label='Validation Accuracy')
plt.axhline(y=test_accuracy, color='r', linestyle='--', label='Test Accuracy')
plt.ylim([0.3,0.9])
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.title('Training and Validation Accuracy')
plt.legend()
plt.grid(True)
plt.show()
Predicting New Images with the above Trained Model¶
In [ ]:
import os
import numpy as np
from tensorflow.keras.preprocessing.image import load_img, img_to_array
import matplotlib.pyplot as plt
def preprocess_image(img_path):
img = load_img(img_path, target_size=(IMG_HEIGHT, IMG_WIDTH))
img_arr = img_to_array(img)
img_arr = np.expand_dims(img_arr, axis=0)
img_arr /= 255.0
return img_arr
img_folder = 'NewTest'
img_paths = [os.path.join(img_folder, fname) for fname in os.listdir(img_folder)]
with_mask_imgs = []
without_mask_imgs = []
for path in img_paths:
try:
img_arr = preprocess_image(path)
pred = model.predict(img_arr)[0][0]
label = 'Without Mask' if pred > 0.5 else 'With Mask'
print(f"Image: {path}, Prediction: {pred:.4f}, Label: {label}")
if label == 'Without Mask':
without_mask_imgs.append(img_arr[0])
else:
with_mask_imgs.append(img_arr[0])
except Exception as e:
print(f"Error with image {path}: {e}")
def display_images(img_list, labels):
fig, axes = plt.subplots(1, len(img_list), figsize=(20, 20))
axes = axes.flatten()
for img, ax, label in zip(img_list, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title(label)
plt.tight_layout()
plt.show()
if with_mask_imgs:
display_images(with_mask_imgs, ['With Mask'] * len(with_mask_imgs))
if without_mask_imgs:
display_images(without_mask_imgs, ['Without Mask'] * len(without_mask_imgs))
1/1 ━━━━━━━━━━━━━━━━━━━━ 2s 2s/step Image: NewTest\178.png, Prediction: 0.0003, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step Image: NewTest\187.png, Prediction: 0.1475, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step Image: NewTest\190.png, Prediction: 0.0275, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step Image: NewTest\219.png, Prediction: 0.7338, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step Image: NewTest\231.png, Prediction: 0.0447, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 67ms/step Image: NewTest\283.png, Prediction: 0.8905, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step Image: NewTest\350.png, Prediction: 0.0165, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step Image: NewTest\351.png, Prediction: 0.0368, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step Image: NewTest\64.png, Prediction: 0.7557, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step Image: NewTest\65.png, Prediction: 0.7777, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 68ms/step Image: NewTest\66.png, Prediction: 0.7451, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 73ms/step Image: NewTest\69.png, Prediction: 0.7593, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 68ms/step Image: NewTest\70.png, Prediction: 0.2197, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step Image: NewTest\72.png, Prediction: 0.7057, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 70ms/step Image: NewTest\face-13.jpg, Prediction: 0.9347, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 73ms/step Image: NewTest\jh.jpg, Prediction: 0.0407, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 71ms/step Image: NewTest\lk.jpg, Prediction: 0.1037, Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000001EFEE74FBF0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step Image: NewTest\uu.jpg, Prediction: 0.2408, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 74ms/step Image: NewTest\with.jpg, Prediction: 0.0367, Label: With Mask
Ex 4.2: Transfer Learning with ResNet50: Model Training, Evaluation, and Results Using TensorFlow¶
In [16]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import ResNet50
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
import json
import os
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 64
EPOCHS = 10
train_dir = 'Train'
validation_dir = 'Validation'
test_dir = 'Test'
datagen = ImageDataGenerator(rescale=1./255)
train_gen = datagen.flow_from_directory(
train_dir,
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
val_gen = datagen.flow_from_directory(
validation_dir,
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_gen = datagen.flow_from_directory(
test_dir,
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary',
shuffle=False
)
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(IMG_HEIGHT, IMG_WIDTH, 3))
base_model.trainable = False
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(256, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(64, activation='relu')(x)
preds = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=preds)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
history = model.fit(
train_gen,
validation_data=val_gen,
epochs=EPOCHS
)
test_loss, test_accuracy = model.evaluate(test_gen, verbose=2)
print(f'Test accuracy of model using transfer learning with ResNet50: {test_accuracy:.4f}')
model_name = 'resnet50_model'
model_path = f'{model_name}.h5'
model.save(model_path)
print(f'Model saved as {model_path}')
history_dict = history.history
history_filename = f'{model_name}_history.json'
with open(history_filename, 'w') as f:
json.dump(history_dict, f)
print(f'Training history saved as {history_filename}')
results = {
'test_loss': test_loss,
'test_accuracy': test_accuracy
}
results_filename = f'{model_name}_results.json'
with open(results_filename, 'w') as f:
json.dump(results, f)
print(f'Test results saved as {results_filename}')
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes. Epoch 1/10
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
50/50 ━━━━━━━━━━━━━━━━━━━━ 90s 1s/step - accuracy: 0.5213 - loss: 0.6926 - val_accuracy: 0.5750 - val_loss: 0.6638 Epoch 2/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 77s 1s/step - accuracy: 0.6393 - loss: 0.6386 - val_accuracy: 0.7469 - val_loss: 0.5980 Epoch 3/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 81s 2s/step - accuracy: 0.7220 - loss: 0.5645 - val_accuracy: 0.5734 - val_loss: 0.7248 Epoch 4/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 80s 1s/step - accuracy: 0.6835 - loss: 0.5753 - val_accuracy: 0.8141 - val_loss: 0.4768 Epoch 5/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 77s 1s/step - accuracy: 0.7950 - loss: 0.4619 - val_accuracy: 0.7906 - val_loss: 0.4861 Epoch 6/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 79s 1s/step - accuracy: 0.8297 - loss: 0.4032 - val_accuracy: 0.8484 - val_loss: 0.3833 Epoch 7/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 75s 1s/step - accuracy: 0.8087 - loss: 0.4301 - val_accuracy: 0.8609 - val_loss: 0.3797 Epoch 8/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 134s 2s/step - accuracy: 0.8644 - loss: 0.3339 - val_accuracy: 0.8359 - val_loss: 0.3751 Epoch 9/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 328s 6s/step - accuracy: 0.8448 - loss: 0.3441 - val_accuracy: 0.8672 - val_loss: 0.3624 Epoch 10/10 50/50 ━━━━━━━━━━━━━━━━━━━━ 261s 5s/step - accuracy: 0.8824 - loss: 0.3098 - val_accuracy: 0.8484 - val_loss: 0.3787 10/10 - 36s - 4s/step - accuracy: 0.8594 - loss: 0.3649
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
Test accuracy of model using transfer learning with ResNet50: 0.8594 Model saved as resnet50_model.h5 Training history saved as resnet50_model_history.json Test results saved as resnet50_model_results.json
In [28]:
import tensorflow as tf
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.metrics import confusion_matrix, classification_report
def evaluate_and_plot_confusion_matrix(model_name):
model = tf.keras.models.load_model(f'{model_name}.h5')
test_loss, test_accuracy = model.evaluate(test_generator, verbose=2)
print(f'{model_name} - Loss: {test_loss}, Accuracy: {test_accuracy:.4f}')
predictions = model.predict(test_generator, verbose=2)
predicted_classes = (predictions > 0.5).astype(int).flatten()
true_classes = test_generator.classes
class_names = list(test_generator.class_indices.keys())
print(classification_report(true_classes, predicted_classes, target_names=class_names))
conf_matrix = confusion_matrix(true_classes, predicted_classes)
plt.figure(figsize=(8, 6))
sns.heatmap(conf_matrix, annot=True, fmt='d', xticklabels=class_names, yticklabels=class_names, cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title(f'Confusion Matrix - {model_name}')
plt.show()
evaluate_and_plot_confusion_matrix('resnet50_model')
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 - 13s - 632ms/step - accuracy: 0.8594 - loss: 0.3649
resnet50_model - Loss: 0.3648945093154907, Accuracy: 0.8594
20/20 - 10s - 504ms/step
precision recall f1-score support
WithMask 0.80 0.96 0.87 320
WithoutMask 0.95 0.76 0.84 320
accuracy 0.86 640
macro avg 0.87 0.86 0.86 640
weighted avg 0.87 0.86 0.86 640
In [20]:
plt.figure(figsize=(12, 6))
plt.plot(history.history['accuracy'], label='Train Accuracy')
plt.plot(history.history['val_accuracy'], label='Validation Accuracy')
plt.axhline(y=test_accuracy, color='r', linestyle='--', label='Test Accuracy')
plt.title('Model accuracy [tranfer learning appraoch using ResNet50]')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.ylim([0,1])
plt.legend(loc='upper left')
plt.savefig(f'{model_name}_accuracy_plot.png')
plt.show()
plt.figure(figsize=(12, 6))
plt.plot(history.history['loss'], label='Train Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.axhline(y=test_loss, color='r', linestyle='--', label='Test Loss')
plt.title('Model loss [tranfer learning appraoch using ResNet50]')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.ylim([0,1])
plt.legend(loc='upper left')
plt.savefig(f'{model_name}_loss_plot.png')
plt.show()
Predicting New Data with a Pretrained ResNet50 Model and Visualizing Results¶
In [19]:
import os
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.preprocessing.image import load_img, img_to_array
IMG_HEIGHT = 128
IMG_WIDTH = 128
images_folder = 'NewTest'
model_name = 'resnet50_model.h5'
model = tf.keras.models.load_model(model_name)
def load_and_preprocess_image(image_path):
img = load_img(image_path, target_size=(IMG_HEIGHT, IMG_WIDTH))
img_array = img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = img_array / 255.0
return img_array
image_paths = [os.path.join(images_folder, img_name) for img_name in os.listdir(images_folder)]
with_mask_images = []
without_mask_images = []
for image_path in image_paths:
try:
img_array = load_and_preprocess_image(image_path)
prediction = model.predict(img_array)
label = 'Without Mask' if prediction > 0.5 else 'With Mask'
print(f"Image: {image_path}, Prediction: {prediction[0][0]:.4f}, Label: {label}")
if label == 'Without Mask':
without_mask_images.append(img_array[0])
else:
with_mask_images.append(img_array[0])
except Exception as e:
print(f"Error loading image {image_path}: {e}")
def plot_images(images_arr, labels):
fig, axes = plt.subplots(1, len(images_arr), figsize=(20, 20))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title(label)
plt.tight_layout()
plt.show()
if with_mask_images:
plot_images(with_mask_images, ['With Mask'] * len(with_mask_images))
if without_mask_images:
plot_images(without_mask_images, ['Without Mask'] * len(without_mask_images))
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 2s 2s/step Image: NewTest\178.png, Prediction: 0.0001, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 64ms/step Image: NewTest\187.png, Prediction: 0.1077, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step Image: NewTest\190.png, Prediction: 0.0109, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step Image: NewTest\219.png, Prediction: 0.8486, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step Image: NewTest\231.png, Prediction: 0.0508, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\283.png, Prediction: 0.9734, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\350.png, Prediction: 0.0130, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 65ms/step Image: NewTest\351.png, Prediction: 0.0289, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step Image: NewTest\64.png, Prediction: 0.9281, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\65.png, Prediction: 0.9119, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step Image: NewTest\66.png, Prediction: 0.8181, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 66ms/step Image: NewTest\69.png, Prediction: 0.9082, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step Image: NewTest\70.png, Prediction: 0.1715, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\72.png, Prediction: 0.9039, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step Image: NewTest\face-13.jpg, Prediction: 0.9778, Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\jh.jpg, Prediction: 0.0114, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 60ms/step Image: NewTest\lk.jpg, Prediction: 0.0570, Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x00000208A5D536A0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 59ms/step Image: NewTest\uu.jpg, Prediction: 0.4360, Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 62ms/step Image: NewTest\with.jpg, Prediction: 0.0200, Label: With Mask
Ex 4.3: Comparing Optimizers: Evaluating Test Accuracy for Various Optimization Algorithms.¶
In [ ]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import ResNet50
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
img_height = 128
img_width = 128
batch_size = 32
datagen = ImageDataGenerator(rescale=1./255)
train_gen = datagen.flow_from_directory(
'Train',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
val_gen = datagen.flow_from_directory(
'Validation',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
test_gen = datagen.flow_from_directory(
'Test',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
optimizers = {
'Adam': tf.keras.optimizers.Adam(),
'RMSprop': tf.keras.optimizers.RMSprop(),
'Adagrad': tf.keras.optimizers.Adagrad(),
'Momentum': tf.keras.optimizers.SGD(momentum=0.9)
}
results = {}
for name, opt in optimizers.items():
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(img_height, img_width, 3))
base_model.trainable = False
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(128, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(64, activation='relu')(x)
pred = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=pred)
model.compile(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])
history = model.fit(
train_gen,
validation_data=val_gen,
epochs=10,
verbose=0
)
test_loss, test_acc = model.evaluate(test_gen, verbose=0)
results[name] = test_acc
print(f'Test accuracy with {name}: {test_acc:.4f}')
plt.figure(figsize=(10, 6))
plt.bar(results.keys(), results.values())
plt.title('Test Accuracy for Different Optimizers')
plt.xlabel('Optimizer')
plt.ylabel('Test Accuracy')
plt.ylim([0.0, 1.0])
plt.show()
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes.
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
Test accuracy with Adam: 0.8875 Test accuracy with RMSprop: 0.8578 Test accuracy with Adagrad: 0.5609 Test accuracy with Momentum: 0.5203
Ex 5.1: Transfer Learning with VGG16: Training, Evaluating, and Saving the Model¶
In [10]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import json
import os
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import VGG16
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
from sklearn.metrics import confusion_matrix, classification_report
import seaborn as sns
img_height = 128
img_width = 128
batch_size = 32
epochs = 10
datagen = ImageDataGenerator(rescale=1./255)
train_gen = datagen.flow_from_directory(
'Train',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
val_gen = datagen.flow_from_directory(
'Validation',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
test_gen = datagen.flow_from_directory(
'Test',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary',
shuffle=False
)
base_model = VGG16(weights='imagenet', include_top=False, input_shape=(img_height, img_width, 3))
base_model.trainable = False
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(256, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(64, activation='relu')(x)
pred = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=pred)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
history = model.fit(
train_gen,
validation_data=val_gen,
epochs=epochs
)
model_name = 'vgg16_model'
model_path = f'{model_name}.h5'
model.save(model_path)
print(f'Model saved as {model_path}')
history_dict = history.history
history_filename = f'{model_name}_history.json'
with open(history_filename, 'w') as f:
json.dump(history_dict, f)
print(f'Training history saved as {history_filename}')
test_loss, test_acc = model.evaluate(test_gen, verbose=2)
print(f'{model_name} - Loss: {test_loss}, Accuracy: {test_acc:.4f}')
results = {
'test_loss': test_loss,
'test_accuracy': test_acc
}
results_filename = f'{model_name}_results.json'
with open(results_filename, 'w') as f:
json.dump(results, f)
print(f'Test results saved as {results_filename}')
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes. Epoch 1/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 108s 1s/step - accuracy: 0.9050 - loss: 0.2521 - val_accuracy: 0.9703 - val_loss: 0.0891 Epoch 2/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 109s 1s/step - accuracy: 0.9848 - loss: 0.0493 - val_accuracy: 0.9719 - val_loss: 0.0791 Epoch 3/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 109s 1s/step - accuracy: 0.9863 - loss: 0.0326 - val_accuracy: 0.9656 - val_loss: 0.0881 Epoch 4/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 112s 1s/step - accuracy: 0.9898 - loss: 0.0299 - val_accuracy: 0.9812 - val_loss: 0.0570 Epoch 5/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 109s 1s/step - accuracy: 0.9887 - loss: 0.0262 - val_accuracy: 0.9781 - val_loss: 0.0726 Epoch 6/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 110s 1s/step - accuracy: 0.9945 - loss: 0.0182 - val_accuracy: 0.9781 - val_loss: 0.0623 Epoch 7/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 109s 1s/step - accuracy: 0.9974 - loss: 0.0093 - val_accuracy: 0.9781 - val_loss: 0.0674 Epoch 8/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 109s 1s/step - accuracy: 0.9966 - loss: 0.0113 - val_accuracy: 0.9812 - val_loss: 0.0591 Epoch 9/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 110s 1s/step - accuracy: 0.9974 - loss: 0.0095 - val_accuracy: 0.9766 - val_loss: 0.0733 Epoch 10/10 100/100 ━━━━━━━━━━━━━━━━━━━━ 107s 1s/step - accuracy: 0.9963 - loss: 0.0118 - val_accuracy: 0.9781 - val_loss: 0.0660
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
Model saved as vgg16_model.h5 Training history saved as vgg16_model_history.json 20/20 - 18s - 900ms/step - accuracy: 0.9797 - loss: 0.0811 vgg16_model - Loss: 0.0810772180557251, Accuracy: 0.9797 Test results saved as vgg16_model_results.json
In [11]:
import tensorflow as tf
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.metrics import confusion_matrix, classification_report
def eval_and_plot_cm(model_name):
model = tf.keras.models.load_model(f'{model_name}.h5')
test_loss, test_acc = model.evaluate(test_gen, verbose=2)
print(f'{model_name} - Loss: {test_loss}, Accuracy: {test_acc:.4f}')
preds = model.predict(test_gen, verbose=2)
pred_classes = (preds > 0.5).astype(int).flatten()
true_classes = test_gen.classes
class_names = list(test_gen.class_indices.keys())
print(classification_report(true_classes, pred_classes, target_names=class_names))
cm = confusion_matrix(true_classes, pred_classes)
plt.figure(figsize=(8, 6))
sns.heatmap(cm, annot=True, fmt='d', xticklabels=class_names, yticklabels=class_names, cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title(f'Confusion Matrix - {model_name}')
plt.show()
model_name = 'vgg16_model'
eval_and_plot_cm(model_name)
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 - 18s - 915ms/step - accuracy: 0.9797 - loss: 0.0811
vgg16_model - Loss: 0.0810772180557251, Accuracy: 0.9797
20/20 - 18s - 880ms/step
precision recall f1-score support
WithMask 0.97 0.99 0.98 320
WithoutMask 0.99 0.97 0.98 320
accuracy 0.98 640
macro avg 0.98 0.98 0.98 640
weighted avg 0.98 0.98 0.98 640
In [14]:
plt.figure(figsize=(12, 6))
plt.plot(history.history['accuracy'], label='Train Accuracy')
plt.plot(history.history['val_accuracy'], label='Validation Accuracy')
plt.axhline(y=test_acc, color='r', linestyle='--', label='Test Accuracy')
plt.title('Model accuracy for VGG16')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.legend(loc='upper left')
plt.savefig(f'{model_name}_accuracy_plot.png')
plt.show()
plt.figure(figsize=(12, 6))
plt.plot(history.history['loss'], label='Train Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.axhline(y=test_loss, color='r', linestyle='--', label='Test Loss')
plt.title('Model loss for VGG16')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend(loc='upper left')
plt.savefig(f'{model_name}_loss_plot.png')
plt.show()
Predicting new data.¶
In [18]:
import os
from tensorflow.keras.preprocessing.image import load_img, img_to_array
images_folder = 'NewTest'
def load_and_preprocess_image(image_path):
img = load_img(image_path, target_size=(IMG_HEIGHT, IMG_WIDTH))
img_array = img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = img_array / 255.0
return img_array
image_paths = [os.path.join(images_folder, img_name) for img_name in os.listdir(images_folder)]
with_mask_images = []
without_mask_images = []
for image_path in image_paths:
try:
img_array = load_and_preprocess_image(image_path)
prediction = model.predict(img_array)
label = 'Without Mask' if prediction > 0.5 else 'With Mask'
print(f"Image: {image_path}, Prediction: {prediction}, Label: {label}")
if label == 'Without Mask':
without_mask_images.append(img_array[0])
else:
with_mask_images.append(img_array[0])
except Exception as e:
print(f"Error loading image {image_path}: {e}")
def plot_images(images_arr, labels):
fig, axes = plt.subplots(1, len(images_arr), figsize=(20, 20))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title(label)
plt.tight_layout()
plt.show()
if with_mask_images:
plot_images(with_mask_images, ['With Mask'] * len(with_mask_images))
if without_mask_images:
plot_images(without_mask_images, ['Without Mask'] * len(without_mask_images))
1/1 ━━━━━━━━━━━━━━━━━━━━ 3s 3s/step Image: NewTest\178.png, Prediction: [[2.015764e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step Image: NewTest\187.png, Prediction: [[7.155299e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\190.png, Prediction: [[2.6951785e-12]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\219.png, Prediction: [[6.520963e-13]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Image: NewTest\231.png, Prediction: [[0.00660834]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\283.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\350.png, Prediction: [[1.2132389e-17]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\351.png, Prediction: [[4.4086096e-10]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step Image: NewTest\64.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Image: NewTest\65.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\66.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\69.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\70.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Image: NewTest\72.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Image: NewTest\face-13.jpg, Prediction: [[0.99978155]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\jh.jpg, Prediction: [[0.9999875]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\lk.jpg, Prediction: [[5.97546e-14]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000001AD7E8762F0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Image: NewTest\uu.jpg, Prediction: [[4.3271176e-10]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Image: NewTest\with.jpg, Prediction: [[7.050154e-10]], Label: With Mask
Ex 6.1: Training and Testing Multiple Pretrained Models for this Image Classification problem¶
This script evaluates various pretrained models (VGG19, ResNet101, InceptionV3, DenseNet121, MobileNetV2) using transfer learning. Each model is trained for 5 epochs, and performance metrics such as accuracy, confusion matrix, and classification report are computed to determine which model performs best for the given image classification task.¶
In [44]:
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import json
import os
import seaborn as sns
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications import VGG19, ResNet101, InceptionV3, DenseNet121, MobileNetV2
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D
from tensorflow.keras.models import Model
from sklearn.metrics import confusion_matrix, classification_report
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 32
EPOCHS = 5
MODELS = {
'vgg19': VGG19,
'resnet101': ResNet101,
'inceptionv3': InceptionV3,
'densenet121': DenseNet121,
'mobilenetv2': MobileNetV2
}
params = {
'vgg19': {'include_top': False},
'resnet101': {'include_top': False},
'inceptionv3': {'include_top': False},
'densenet121': {'include_top': False},
'mobilenetv2': {'include_top': False}
}
datagen = ImageDataGenerator(rescale=1./255)
train_gen = datagen.flow_from_directory(
'Train',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
val_gen = datagen.flow_from_directory(
'Validation',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary'
)
test_gen = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary',
shuffle=False
)
def create_model(model_name):
model_class = MODELS[model_name]
base_params = params[model_name]
base_model = model_class(weights='imagenet', input_shape=(IMG_HEIGHT, IMG_WIDTH, 3), **base_params)
base_model.trainable = False
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(256, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(64, activation='relu')(x)
preds = Dense(1, activation='sigmoid')(x)
model = Model(inputs=base_model.input, outputs=preds)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
return model
def train_and_eval_model(model_name):
model = create_model(model_name)
history = model.fit(
train_gen,
validation_data=val_gen,
epochs=EPOCHS
)
model_path = f'{model_name}_model.h5'
model.save(model_path)
print(f'{model_name} model saved as {model_path}')
history_dict = history.history
history_filename = f'{model_name}_history.json'
with open(history_filename, 'w') as f:
json.dump(history_dict, f)
print(f'Training history saved as {history_filename}')
test_loss, test_acc = model.evaluate(test_gen, verbose=2)
print(f'{model_name} - Loss: {test_loss}, Accuracy: {test_acc:.4f}')
results = {
'test_loss': test_loss,
'test_accuracy': test_acc
}
results_filename = f'{model_name}_results.json'
with open(results_filename, 'w') as f:
json.dump(results, f)
print(f'Test results saved as {results_filename}')
preds = model.predict(test_gen, verbose=2)
pred_classes = (preds > 0.5).astype(int).flatten()
true_classes = test_gen.classes
class_names = list(test_gen.class_indices.keys())
print(classification_report(true_classes, pred_classes, target_names=class_names))
cm = confusion_matrix(true_classes, pred_classes)
return test_acc, cm, class_names
cms = {}
accs = {}
for model_name in MODELS.keys():
acc, cm, class_names = train_and_eval_model(model_name)
accs[model_name] = acc
cms[model_name] = (cm, class_names)
Found 3200 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes. Epoch 1/5
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
100/100 ━━━━━━━━━━━━━━━━━━━━ 148s 1s/step - accuracy: 0.8448 - loss: 0.3408 - val_accuracy: 0.9609 - val_loss: 0.0944 Epoch 2/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 143s 1s/step - accuracy: 0.9687 - loss: 0.0803 - val_accuracy: 0.9750 - val_loss: 0.0672 Epoch 3/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 142s 1s/step - accuracy: 0.9761 - loss: 0.0598 - val_accuracy: 0.9766 - val_loss: 0.0651 Epoch 4/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 142s 1s/step - accuracy: 0.9819 - loss: 0.0529 - val_accuracy: 0.9688 - val_loss: 0.0727 Epoch 5/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 150s 1s/step - accuracy: 0.9880 - loss: 0.0332 - val_accuracy: 0.9719 - val_loss: 0.0833
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
vgg19 model saved as vgg19_model.h5
Training history saved as vgg19_history.json
20/20 - 23s - 1s/step - accuracy: 0.9688 - loss: 0.1004
vgg19 - Loss: 0.10040515661239624, Accuracy: 0.9688
Test results saved as vgg19_results.json
20/20 - 25s - 1s/step
precision recall f1-score support
WithMask 0.96 0.98 0.97 320
WithoutMask 0.98 0.96 0.97 320
accuracy 0.97 640
macro avg 0.97 0.97 0.97 640
weighted avg 0.97 0.97 0.97 640
Epoch 1/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 108s 919ms/step - accuracy: 0.5625 - loss: 0.6800 - val_accuracy: 0.6297 - val_loss: 0.6304
Epoch 2/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 89s 874ms/step - accuracy: 0.6918 - loss: 0.5821 - val_accuracy: 0.7734 - val_loss: 0.5087
Epoch 3/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 93s 901ms/step - accuracy: 0.7725 - loss: 0.4771 - val_accuracy: 0.7953 - val_loss: 0.4618
Epoch 4/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 90s 874ms/step - accuracy: 0.8093 - loss: 0.4199 - val_accuracy: 0.8078 - val_loss: 0.4356
Epoch 5/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 87s 841ms/step - accuracy: 0.8213 - loss: 0.3968 - val_accuracy: 0.8188 - val_loss: 0.4073
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
resnet101 model saved as resnet101_model.h5
Training history saved as resnet101_history.json
20/20 - 16s - 800ms/step - accuracy: 0.8266 - loss: 0.4058
resnet101 - Loss: 0.40579086542129517, Accuracy: 0.8266
Test results saved as resnet101_results.json
20/20 - 19s - 954ms/step
precision recall f1-score support
WithMask 0.85 0.79 0.82 320
WithoutMask 0.80 0.86 0.83 320
accuracy 0.83 640
macro avg 0.83 0.83 0.83 640
weighted avg 0.83 0.83 0.83 640
Epoch 1/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 64s 543ms/step - accuracy: 0.9289 - loss: 0.1561 - val_accuracy: 0.9797 - val_loss: 0.0713
Epoch 2/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 50s 477ms/step - accuracy: 0.9902 - loss: 0.0224 - val_accuracy: 0.9875 - val_loss: 0.0801
Epoch 3/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 50s 467ms/step - accuracy: 0.9933 - loss: 0.0127 - val_accuracy: 0.9797 - val_loss: 0.0685
Epoch 4/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 51s 490ms/step - accuracy: 0.9966 - loss: 0.0096 - val_accuracy: 0.9812 - val_loss: 0.0728
Epoch 5/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 50s 467ms/step - accuracy: 1.0000 - loss: 0.0022 - val_accuracy: 0.9844 - val_loss: 0.0856
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
inceptionv3 model saved as inceptionv3_model.h5
Training history saved as inceptionv3_history.json
20/20 - 8s - 379ms/step - accuracy: 0.9953 - loss: 0.0098
inceptionv3 - Loss: 0.009802782908082008, Accuracy: 0.9953
Test results saved as inceptionv3_results.json
20/20 - 7s - 330ms/step
precision recall f1-score support
WithMask 0.99 1.00 1.00 320
WithoutMask 1.00 0.99 1.00 320
accuracy 1.00 640
macro avg 1.00 1.00 1.00 640
weighted avg 1.00 1.00 1.00 640
Epoch 1/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 78s 611ms/step - accuracy: 0.9513 - loss: 0.1023 - val_accuracy: 0.9937 - val_loss: 0.0176
Epoch 2/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 58s 551ms/step - accuracy: 0.9968 - loss: 0.0140 - val_accuracy: 0.9937 - val_loss: 0.0128
Epoch 3/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 58s 548ms/step - accuracy: 0.9950 - loss: 0.0073 - val_accuracy: 0.9953 - val_loss: 0.0185
Epoch 4/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 58s 549ms/step - accuracy: 0.9994 - loss: 0.0013 - val_accuracy: 0.9953 - val_loss: 0.0183
Epoch 5/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 59s 564ms/step - accuracy: 0.9975 - loss: 0.0058 - val_accuracy: 0.9937 - val_loss: 0.0322
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
densenet121 model saved as densenet121_model.h5
Training history saved as densenet121_history.json
20/20 - 11s - 541ms/step - accuracy: 0.9922 - loss: 0.0359
densenet121 - Loss: 0.035893939435482025, Accuracy: 0.9922
Test results saved as densenet121_results.json
20/20 - 13s - 652ms/step
precision recall f1-score support
WithMask 0.98 1.00 0.99 320
WithoutMask 1.00 0.98 0.99 320
accuracy 0.99 640
macro avg 0.99 0.99 0.99 640
weighted avg 0.99 0.99 0.99 640
Epoch 1/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 57s 489ms/step - accuracy: 0.9375 - loss: 0.1423 - val_accuracy: 0.9875 - val_loss: 0.0395
Epoch 2/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 49s 462ms/step - accuracy: 0.9941 - loss: 0.0151 - val_accuracy: 0.9844 - val_loss: 0.0535
Epoch 3/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 49s 463ms/step - accuracy: 0.9969 - loss: 0.0075 - val_accuracy: 0.9797 - val_loss: 0.0926
Epoch 4/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 48s 455ms/step - accuracy: 0.9978 - loss: 0.0090 - val_accuracy: 0.9859 - val_loss: 0.0613
Epoch 5/5
100/100 ━━━━━━━━━━━━━━━━━━━━ 49s 459ms/step - accuracy: 0.9971 - loss: 0.0079 - val_accuracy: 0.9844 - val_loss: 0.0859
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
mobilenetv2 model saved as mobilenetv2_model.h5
Training history saved as mobilenetv2_history.json
20/20 - 7s - 371ms/step - accuracy: 0.9859 - loss: 0.0872
mobilenetv2 - Loss: 0.0871998518705368, Accuracy: 0.9859
Test results saved as mobilenetv2_results.json
20/20 - 5s - 231ms/step
precision recall f1-score support
WithMask 0.97 1.00 0.99 320
WithoutMask 1.00 0.97 0.99 320
accuracy 0.99 640
macro avg 0.99 0.99 0.99 640
weighted avg 0.99 0.99 0.99 640
In [45]:
plt.figure(figsize=(20, 8))
for i, (model_name, (cm, class_names)) in enumerate(confusion_matrices.items(), start=1):
plt.subplot(1, len(MODELS), i)
sns.heatmap(cm, annot=True, fmt='d', xticklabels=class_names, yticklabels=class_names, cmap='Blues')
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title(f'Confusion Matrix - {model_name}')
plt.tight_layout()
plt.savefig('confusion_matrices.png')
plt.show()
plt.figure(figsize=(10, 6))
plt.bar(test_accuracies.keys(), test_accuracies.values(), color='skyblue')
plt.xlabel('Model')
plt.ylabel('Test Accuracy')
plt.title('Test Accuracy vs Model')
plt.xticks(rotation=45)
plt.ylim(0, 1)
plt.tight_layout()
plt.savefig('test_accuracies.png')
plt.show()
In [ ]:
import os
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.models import load_model
from sklearn.metrics import classification_report, roc_curve, auc
IMG_HEIGHT = 128
IMG_WIDTH = 128
BATCH_SIZE = 32
MODEL_NAMES = ['vgg19_model.h5', 'resnet101_model.h5', 'inceptionv3_model.h5', 'densenet121_model.h5', 'mobilenetv2_model.h5']
datagen = ImageDataGenerator(rescale=1./255)
test_generator = datagen.flow_from_directory(
'Test',
target_size=(IMG_HEIGHT, IMG_WIDTH),
batch_size=BATCH_SIZE,
class_mode='binary',
shuffle=False
)
performance_metrics = {}
for model_name in MODEL_NAMES:
model = load_model(model_name)
y_true = test_generator.classes
y_pred_probs = model.predict(test_generator)
y_pred = (y_pred_probs > 0.5).astype(int)
report = classification_report(y_true, y_pred, output_dict=True)
precision = report['1']['precision']
recall = report['1']['recall']
f1_score = report['1']['f1-score']
fpr, tpr, _ = roc_curve(y_true, y_pred_probs)
roc_auc = auc(fpr, tpr)
model_key = model_name.split('_')[0]
performance_metrics[model_key] = {
'precision': precision,
'recall': recall,
'f1_score': f1_score,
'roc_auc': roc_auc,
'fpr': fpr,
'tpr': tpr,
}
fig, axs = plt.subplots(2, 2, figsize=(15, 15))
# Precision
axs[0, 0].bar(performance_metrics.keys(), [metrics['precision'] for metrics in performance_metrics.values()], color='skyblue')
axs[0, 0].set_title('Precision')
axs[0, 0].set_ylabel('Precision')
axs[0, 0].set_ylim(0, 1)
# Recall
axs[0, 1].bar(performance_metrics.keys(), [metrics['recall'] for metrics in performance_metrics.values()], color='lightgreen')
axs[0, 1].set_title('Recall')
axs[0, 1].set_ylabel('Recall')
axs[0, 1].set_ylim(0, 1)
# F1-Score
axs[1, 0].bar(performance_metrics.keys(), [metrics['f1_score'] for metrics in performance_metrics.values()], color='salmon')
axs[1, 0].set_title('F1-Score')
axs[1, 0].set_ylabel('F1-Score')
axs[1, 0].set_ylim(0, 1)
# ROC Curves
for model_name, metrics in performance_metrics.items():
axs[1, 1].plot(metrics['fpr'], metrics['tpr'], label=f'{model_name} (AUC = {metrics["roc_auc"]:.2f})')
axs[1, 1].plot([0, 1], [0, 1], 'k--')
axs[1, 1].set_title('ROC Curves')
axs[1, 1].set_xlabel('False Positive Rate')
axs[1, 1].set_ylabel('True Positive Rate')
axs[1, 1].legend(loc='lower right')
plt.tight_layout()
plt.savefig('performance_metrics.png')
plt.show()
Found 640 images belonging to 2 classes.
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model. C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:121: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
20/20 ━━━━━━━━━━━━━━━━━━━━ 24s 1s/step
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 ━━━━━━━━━━━━━━━━━━━━ 18s 726ms/step
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 ━━━━━━━━━━━━━━━━━━━━ 9s 355ms/step
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 ━━━━━━━━━━━━━━━━━━━━ 13s 472ms/step
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
20/20 ━━━━━━━━━━━━━━━━━━━━ 5s 193ms/step
In [3]:
import os
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.preprocessing.image import load_img, img_to_array
from tensorflow.keras.models import load_model
IMG_HEIGHT = 128
IMG_WIDTH = 128
images_folder = 'NewTest'
models_folder = './'
def load_and_preprocess_image(image_path):
img = load_img(image_path, target_size=(IMG_HEIGHT, IMG_WIDTH))
img_array = img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = img_array / 255.0
return img_array
def plot_images(images_arr, labels, model_name):
fig, axes = plt.subplots(1, len(images_arr), figsize=(20, 4))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title(label)
plt.tight_layout()
plt.suptitle(f'Predictions by {model_name}', size=16)
plt.show()
model_names = [ 'inceptionv3_model.h5', 'densenet121_model.h5', 'vgg19_model.h5','mobilenetv2_model.h5', 'resnet101_model.h5']
image_paths = [os.path.join(images_folder, img_name) for img_name in os.listdir(images_folder)]
for model_name in model_names:
model_path = os.path.join(models_folder, model_name)
model = load_model(model_path)
with_mask_images = []
without_mask_images = []
for image_path in image_paths:
try:
img_array = load_and_preprocess_image(image_path)
prediction = model.predict(img_array)
label = 'Without Mask' if prediction > 0.5 else 'With Mask'
print(f"Model: {model_name}, Image: {image_path}, Prediction: {prediction}, Label: {label}")
if label == 'Without Mask':
without_mask_images.append(img_array[0])
else:
with_mask_images.append(img_array[0])
except Exception as e:
print(f"Error loading image {image_path}: {e}")
if with_mask_images:
plot_images(with_mask_images, ['With Mask'] * len(with_mask_images), model_name)
if without_mask_images:
plot_images(without_mask_images, ['Without Mask'] * len(without_mask_images), model_name)
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 2s 2s/step Model: inceptionv3_model.h5, Image: NewTest\178.png, Prediction: [[4.866569e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step Model: inceptionv3_model.h5, Image: NewTest\187.png, Prediction: [[1.5346592e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step Model: inceptionv3_model.h5, Image: NewTest\190.png, Prediction: [[1.4227006e-10]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step Model: inceptionv3_model.h5, Image: NewTest\219.png, Prediction: [[2.7643983e-11]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step Model: inceptionv3_model.h5, Image: NewTest\231.png, Prediction: [[4.5847122e-05]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step Model: inceptionv3_model.h5, Image: NewTest\283.png, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step Model: inceptionv3_model.h5, Image: NewTest\350.png, Prediction: [[1.7430943e-12]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step Model: inceptionv3_model.h5, Image: NewTest\351.png, Prediction: [[2.5000173e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Model: inceptionv3_model.h5, Image: NewTest\69.png, Prediction: [[0.99999064]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step Model: inceptionv3_model.h5, Image: NewTest\72.png, Prediction: [[0.9999989]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Model: inceptionv3_model.h5, Image: NewTest\download (1).jpg, Prediction: [[2.3009818e-07]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step Model: inceptionv3_model.h5, Image: NewTest\download (2).jpg, Prediction: [[0.00128012]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step Model: inceptionv3_model.h5, Image: NewTest\download (3).jpg, Prediction: [[9.693503e-11]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 48ms/step Model: inceptionv3_model.h5, Image: NewTest\download (4).jpg, Prediction: [[0.9999996]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: inceptionv3_model.h5, Image: NewTest\download (6).jpg, Prediction: [[0.9997336]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 51ms/step Model: inceptionv3_model.h5, Image: NewTest\download (7).jpg, Prediction: [[0.999875]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step Model: inceptionv3_model.h5, Image: NewTest\download.jpg, Prediction: [[8.643129e-05]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Model: inceptionv3_model.h5, Image: NewTest\face-13.jpg, Prediction: [[0.9998884]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step Model: inceptionv3_model.h5, Image: NewTest\images (2).jpg, Prediction: [[0.9999999]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Model: inceptionv3_model.h5, Image: NewTest\images (3).jpg, Prediction: [[1.1928866e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step Model: inceptionv3_model.h5, Image: NewTest\images (4).jpg, Prediction: [[1.]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 53ms/step Model: inceptionv3_model.h5, Image: NewTest\images.jpg, Prediction: [[0.9999978]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Model: inceptionv3_model.h5, Image: NewTest\jh.jpg, Prediction: [[0.99891645]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 52ms/step Model: inceptionv3_model.h5, Image: NewTest\kjk.jpg, Prediction: [[1.2814795e-05]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Model: inceptionv3_model.h5, Image: NewTest\lk.jpg, Prediction: [[6.8766604e-11]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000002409D040FE0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 50ms/step Model: inceptionv3_model.h5, Image: NewTest\uu.jpg, Prediction: [[1.3613885e-07]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 49ms/step Model: inceptionv3_model.h5, Image: NewTest\with.jpg, Prediction: [[3.0111744e-07]], Label: With Mask
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 3s 3s/step Model: densenet121_model.h5, Image: NewTest\178.png, Prediction: [[4.049454e-12]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 76ms/step Model: densenet121_model.h5, Image: NewTest\187.png, Prediction: [[1.6674944e-15]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 77ms/step Model: densenet121_model.h5, Image: NewTest\190.png, Prediction: [[5.676101e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\219.png, Prediction: [[1.3603387e-11]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\231.png, Prediction: [[6.5472124e-12]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 89ms/step Model: densenet121_model.h5, Image: NewTest\283.png, Prediction: [[0.9999995]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\350.png, Prediction: [[1.3997115e-14]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\351.png, Prediction: [[4.3135786e-14]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\69.png, Prediction: [[0.99999547]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\72.png, Prediction: [[0.99989396]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 104ms/step Model: densenet121_model.h5, Image: NewTest\download (1).jpg, Prediction: [[3.7740623e-11]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\download (2).jpg, Prediction: [[1.0595586e-06]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Model: densenet121_model.h5, Image: NewTest\download (3).jpg, Prediction: [[1.367655e-14]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 88ms/step Model: densenet121_model.h5, Image: NewTest\download (4).jpg, Prediction: [[0.99999464]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\download (6).jpg, Prediction: [[0.91314995]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 84ms/step Model: densenet121_model.h5, Image: NewTest\download (7).jpg, Prediction: [[0.999957]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 81ms/step Model: densenet121_model.h5, Image: NewTest\download.jpg, Prediction: [[7.330307e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 82ms/step Model: densenet121_model.h5, Image: NewTest\face-13.jpg, Prediction: [[0.9998376]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 80ms/step Model: densenet121_model.h5, Image: NewTest\images (2).jpg, Prediction: [[0.9997648]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 81ms/step Model: densenet121_model.h5, Image: NewTest\images (3).jpg, Prediction: [[2.8275544e-14]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Model: densenet121_model.h5, Image: NewTest\images (4).jpg, Prediction: [[0.999999]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 104ms/step Model: densenet121_model.h5, Image: NewTest\images.jpg, Prediction: [[0.99924994]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\jh.jpg, Prediction: [[0.998495]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Model: densenet121_model.h5, Image: NewTest\kjk.jpg, Prediction: [[1.407495e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 80ms/step Model: densenet121_model.h5, Image: NewTest\lk.jpg, Prediction: [[5.2334857e-18]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000002409B3643B0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\uu.jpg, Prediction: [[2.190711e-10]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 78ms/step Model: densenet121_model.h5, Image: NewTest\with.jpg, Prediction: [[4.7710963e-12]], Label: With Mask
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 345ms/step Model: vgg19_model.h5, Image: NewTest\178.png, Prediction: [[0.7063502]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 115ms/step Model: vgg19_model.h5, Image: NewTest\187.png, Prediction: [[0.00455936]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 116ms/step Model: vgg19_model.h5, Image: NewTest\190.png, Prediction: [[0.00591513]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 138ms/step Model: vgg19_model.h5, Image: NewTest\219.png, Prediction: [[0.00491378]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 126ms/step Model: vgg19_model.h5, Image: NewTest\231.png, Prediction: [[0.00137972]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 109ms/step Model: vgg19_model.h5, Image: NewTest\283.png, Prediction: [[0.9998327]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 133ms/step Model: vgg19_model.h5, Image: NewTest\350.png, Prediction: [[0.00395094]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 116ms/step Model: vgg19_model.h5, Image: NewTest\351.png, Prediction: [[0.02559365]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 128ms/step Model: vgg19_model.h5, Image: NewTest\69.png, Prediction: [[0.9993326]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 117ms/step Model: vgg19_model.h5, Image: NewTest\72.png, Prediction: [[0.9999274]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 115ms/step Model: vgg19_model.h5, Image: NewTest\download (1).jpg, Prediction: [[0.00785471]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 125ms/step Model: vgg19_model.h5, Image: NewTest\download (2).jpg, Prediction: [[0.00179739]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 112ms/step Model: vgg19_model.h5, Image: NewTest\download (3).jpg, Prediction: [[0.00506361]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 132ms/step Model: vgg19_model.h5, Image: NewTest\download (4).jpg, Prediction: [[0.9946115]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 129ms/step Model: vgg19_model.h5, Image: NewTest\download (6).jpg, Prediction: [[0.23090325]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 130ms/step Model: vgg19_model.h5, Image: NewTest\download (7).jpg, Prediction: [[0.9902496]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 133ms/step Model: vgg19_model.h5, Image: NewTest\download.jpg, Prediction: [[0.06610495]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 133ms/step Model: vgg19_model.h5, Image: NewTest\face-13.jpg, Prediction: [[0.99903226]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step Model: vgg19_model.h5, Image: NewTest\images (2).jpg, Prediction: [[0.9911857]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 112ms/step Model: vgg19_model.h5, Image: NewTest\images (3).jpg, Prediction: [[4.278071e-05]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 165ms/step Model: vgg19_model.h5, Image: NewTest\images (4).jpg, Prediction: [[0.99855036]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 114ms/step Model: vgg19_model.h5, Image: NewTest\images.jpg, Prediction: [[0.9611288]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 130ms/step Model: vgg19_model.h5, Image: NewTest\jh.jpg, Prediction: [[0.998042]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 116ms/step Model: vgg19_model.h5, Image: NewTest\kjk.jpg, Prediction: [[0.00320621]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 127ms/step Model: vgg19_model.h5, Image: NewTest\lk.jpg, Prediction: [[0.08308721]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000002414E448450> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 150ms/step Model: vgg19_model.h5, Image: NewTest\uu.jpg, Prediction: [[0.00821439]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 116ms/step Model: vgg19_model.h5, Image: NewTest\with.jpg, Prediction: [[0.01482262]], Label: With Mask
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 1s 1s/step Model: mobilenetv2_model.h5, Image: NewTest\178.png, Prediction: [[3.8931325e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 32ms/step Model: mobilenetv2_model.h5, Image: NewTest\187.png, Prediction: [[1.9640814e-10]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 34ms/step Model: mobilenetv2_model.h5, Image: NewTest\190.png, Prediction: [[1.1401611e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 35ms/step Model: mobilenetv2_model.h5, Image: NewTest\219.png, Prediction: [[2.033661e-06]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\231.png, Prediction: [[1.3559447e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\283.png, Prediction: [[0.9999933]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\350.png, Prediction: [[8.678426e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\351.png, Prediction: [[6.6755015e-09]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 43ms/step Model: mobilenetv2_model.h5, Image: NewTest\69.png, Prediction: [[0.9999999]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\72.png, Prediction: [[0.99999934]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (1).jpg, Prediction: [[3.3248477e-06]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (2).jpg, Prediction: [[2.0523949e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (3).jpg, Prediction: [[2.550458e-06]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 37ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (4).jpg, Prediction: [[0.99997985]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 38ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (6).jpg, Prediction: [[0.27105913]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\download (7).jpg, Prediction: [[0.9999996]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 39ms/step Model: mobilenetv2_model.h5, Image: NewTest\download.jpg, Prediction: [[0.00508273]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\face-13.jpg, Prediction: [[0.9998944]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 46ms/step Model: mobilenetv2_model.h5, Image: NewTest\images (2).jpg, Prediction: [[0.542467]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\images (3).jpg, Prediction: [[8.846442e-08]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\images (4).jpg, Prediction: [[0.9999997]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\images.jpg, Prediction: [[0.99934]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\jh.jpg, Prediction: [[0.9738937]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 39ms/step Model: mobilenetv2_model.h5, Image: NewTest\kjk.jpg, Prediction: [[1.5404297e-05]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\lk.jpg, Prediction: [[5.7024226e-07]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000002415529C3B0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 47ms/step Model: mobilenetv2_model.h5, Image: NewTest\uu.jpg, Prediction: [[1.3776989e-07]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 31ms/step Model: mobilenetv2_model.h5, Image: NewTest\with.jpg, Prediction: [[7.2912226e-05]], Label: With Mask
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
1/1 ━━━━━━━━━━━━━━━━━━━━ 3s 3s/step Model: resnet101_model.h5, Image: NewTest\178.png, Prediction: [[0.00427103]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 85ms/step Model: resnet101_model.h5, Image: NewTest\187.png, Prediction: [[0.07194897]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 101ms/step Model: resnet101_model.h5, Image: NewTest\190.png, Prediction: [[0.04231601]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 82ms/step Model: resnet101_model.h5, Image: NewTest\219.png, Prediction: [[0.09886786]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 108ms/step Model: resnet101_model.h5, Image: NewTest\231.png, Prediction: [[0.0526559]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 81ms/step Model: resnet101_model.h5, Image: NewTest\283.png, Prediction: [[0.9242979]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 105ms/step Model: resnet101_model.h5, Image: NewTest\350.png, Prediction: [[0.00505993]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 90ms/step Model: resnet101_model.h5, Image: NewTest\351.png, Prediction: [[0.17502335]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Model: resnet101_model.h5, Image: NewTest\69.png, Prediction: [[0.90762043]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 96ms/step Model: resnet101_model.h5, Image: NewTest\72.png, Prediction: [[0.9730585]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 81ms/step Model: resnet101_model.h5, Image: NewTest\download (1).jpg, Prediction: [[0.24956013]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 104ms/step Model: resnet101_model.h5, Image: NewTest\download (2).jpg, Prediction: [[0.13656315]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 83ms/step Model: resnet101_model.h5, Image: NewTest\download (3).jpg, Prediction: [[0.19278844]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 106ms/step Model: resnet101_model.h5, Image: NewTest\download (4).jpg, Prediction: [[0.04141852]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 98ms/step Model: resnet101_model.h5, Image: NewTest\download (6).jpg, Prediction: [[0.00810113]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 100ms/step Model: resnet101_model.h5, Image: NewTest\download (7).jpg, Prediction: [[0.18396632]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step Model: resnet101_model.h5, Image: NewTest\download.jpg, Prediction: [[0.02391559]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step Model: resnet101_model.h5, Image: NewTest\face-13.jpg, Prediction: [[0.8726716]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 83ms/step Model: resnet101_model.h5, Image: NewTest\images (2).jpg, Prediction: [[0.01507649]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 102ms/step Model: resnet101_model.h5, Image: NewTest\images (3).jpg, Prediction: [[0.18045199]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 93ms/step Model: resnet101_model.h5, Image: NewTest\images (4).jpg, Prediction: [[0.0374297]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 113ms/step Model: resnet101_model.h5, Image: NewTest\images.jpg, Prediction: [[0.55393183]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 99ms/step Model: resnet101_model.h5, Image: NewTest\jh.jpg, Prediction: [[0.05456086]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 104ms/step Model: resnet101_model.h5, Image: NewTest\kjk.jpg, Prediction: [[0.0306377]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 91ms/step Model: resnet101_model.h5, Image: NewTest\lk.jpg, Prediction: [[0.1273285]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x000002414B66D7B0> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 97ms/step Model: resnet101_model.h5, Image: NewTest\uu.jpg, Prediction: [[0.3339463]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 85ms/step Model: resnet101_model.h5, Image: NewTest\with.jpg, Prediction: [[0.00649158]], Label: With Mask
Ex 7.1: Using Active learning and verifying its effect on models performance.¶
Defining the methods and loading data¶
In [40]:
import tensorflow as tf
import numpy as np
import os
import shutil
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.models import load_model
img_height = 128
img_width = 128
batch_size = 32
epochs = 10
samples_to_label = 100
data_gen = ImageDataGenerator(rescale=1./255)
train_data = data_gen.flow_from_directory(
'Train',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
val_data = data_gen.flow_from_directory(
'Validation',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
test_data = data_gen.flow_from_directory(
'Test',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary',
shuffle=False
)
unlabeled_data = data_gen.flow_from_directory(
'Unlabeled',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode=None,
shuffle=True
)
def active_learning_iteration(model_name, model):
model_path = os.path.join('./', model_name)
model = load_model(model_path)
predictions = model.predict(unlabeled_data, verbose=1)
uncertainties = np.abs(predictions - 0.5).flatten()
uncertain_indices = np.argsort(uncertainties)[:samples_to_label]
uncertain_files = np.array(unlabeled_data.filenames)[uncertain_indices]
filenames = [os.path.basename(f) for f in uncertain_files]
print(filenames)
return model, filenames
def move_images_to_file(files_to_move):
unlabeled_dir = 'Unlabeled\\Unlabeled'
train_dir = 'Train'
mask_dir = os.path.join(train_dir, 'withMask')
no_mask_dir = os.path.join(train_dir, 'withoutMask')
os.makedirs(mask_dir, exist_ok=True)
os.makedirs(no_mask_dir, exist_ok=True)
for filename in files_to_move:
file_path = os.path.join(unlabeled_dir, filename)
if os.path.isfile(file_path):
if filename.startswith('Augmented'):
shutil.move(file_path, os.path.join(mask_dir, filename))
elif filename[0].isdigit():
shutil.move(file_path, os.path.join(no_mask_dir, filename))
else:
print(f"File not found: {file_path}")
print("Files have been moved to the respective directories.")
def retrain_model(model_name, model):
model_path = os.path.join('./', model_name)
model = load_model(model_path)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
train_data = data_gen.flow_from_directory(
'Train',
target_size=(img_height, img_width),
batch_size=batch_size,
class_mode='binary'
)
history = model.fit(train_data, validation_data=val_data, epochs=5)
return model
Found 3099 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 640 images belonging to 2 classes. Found 100 images belonging to 1 classes.
Iteration 1¶
In [36]:
model = None
model, files = active_learning_iteration('resnet101_model.h5', model)
print()
print()
move_images_to_file(files)
print()
print()
model = train_model('resnet101_model.h5', model)
model_name = 'resnet101_model_iteration_1.h5'
model_path = os.path.join('./', model_name)
model.save(model_path)
print(f'Model saved as {model_path}')
history_data = history.history
history_file = f'{model_name}_history.json'
with open(history_file, 'w') as f:
json.dump(history_data, f)
print(f'Training history saved as {history_file}')
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Active Learning Iteration with model: resnet101_model.h5
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
10/10 ━━━━━━━━━━━━━━━━━━━━ 14s 1s/step ['Augmented_188_1011922.png', '2281.png', '2342.png', 'Augmented_195_9835905.png', 'Augmented_190_2672112.png', 'Augmented_142_9403771.png', 'Augmented_200_9667529.png', 'Augmented_127_99.png', '2116.png', 'Augmented_155_8677387.png', '2397.png', 'Augmented_188_1033850.png', '2503.png', '2148.png', 'Augmented_122_9885756.png', 'Augmented_129_7974032.png', 'Augmented_164_234928.png', 'Augmented_123_1480540.png', 'Augmented_203_3327352.png', '2305.png', '2310.png', 'Augmented_130_934402.png', 'Augmented_153_6511400.png', 'Augmented_167_7768054.png', '2322.png', '2358.png', 'Augmented_166_3048399.png', '2390.png', '2339.png', '2147.png', '2303.png', '2506.png', '2367.png', '2487.png', 'Augmented_122_7836976.png', 'Augmented_201_1670634.png', '2362.png', 'Augmented_841_4526145.png', '2492.png', '2369.png', '2505.png', 'Augmented_154_4469620.png', '2302.png', '2110.png', '2323.png', '2298.png', 'Augmented_121_9997608.png', 'Augmented_157_9032932.png', '2144.png', 'Augmented_190_3242422.png', '2328.png', 'Augmented_131_6721263.png', '2375.png', '2150.png', 'Augmented_130_8469519.png', 'Augmented_132_8254415.png', 'Augmented_186_5808531.png', 'Augmented_128_5525629.png', '2355.png', 'Augmented_127_8239705.png', 'Augmented_137_1096551.png', '2378.png', 'Augmented_129_8420896.png', 'Augmented_132_8489012.png', 'Augmented_187_1419015.png', '2314.png', 'Augmented_162_7247410.png', 'Augmented_205_7294700.png', '2179.png', 'Augmented_168_870163.png', '2317.png', 'Augmented_144_2156065.png', '2514.png', '2371.png', 'Augmented_203_528857.png', 'Augmented_132_5144913.png', '2478.png', '2301.png', '2072.png', 'Augmented_202_3408530.png', 'Augmented_187_9534370.png', 'Augmented_143_1838298.png', 'Augmented_153_928455.png', 'Augmented_154_2609759.png', 'Augmented_129_1593165.png', 'Augmented_185_649926.png', 'Augmented_162_4190253.png', 'Augmented_130_1378196.png', 'Augmented_196_454378.png', '2350.png', '2315.png', 'Augmented_131_7359198.png', '2289.png', 'Augmented_184_6240562.png', 'Augmented_125_9782701.png', 'Augmented_153_3441598.png', '2312.png', 'Augmented_152_6932890.png', 'Augmented_159_9657147.png', '2349.png'] Selected files have been successfully moved to their respective directories.
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Found 2999 images belonging to 2 classes. Epoch 1/5 94/94 ━━━━━━━━━━━━━━━━━━━━ 102s 926ms/step - accuracy: 0.8252 - loss: 0.3881 - val_accuracy: 0.8188 - val_loss: 0.4066 Epoch 2/5 94/94 ━━━━━━━━━━━━━━━━━━━━ 87s 884ms/step - accuracy: 0.8390 - loss: 0.3639 - val_accuracy: 0.7469 - val_loss: 0.5099 Epoch 3/5 94/94 ━━━━━━━━━━━━━━━━━━━━ 86s 886ms/step - accuracy: 0.8531 - loss: 0.3623 - val_accuracy: 0.8203 - val_loss: 0.3988 Epoch 4/5 94/94 ━━━━━━━━━━━━━━━━━━━━ 85s 868ms/step - accuracy: 0.8637 - loss: 0.3237 - val_accuracy: 0.8375 - val_loss: 0.3677 Epoch 5/5 94/94 ━━━━━━━━━━━━━━━━━━━━ 86s 879ms/step - accuracy: 0.8730 - loss: 0.3086 - val_accuracy: 0.7797 - val_loss: 0.5303
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
Model saved as ./resnet101_model_iteration_1.h5 Training history saved as resnet101_model_iteration_1.h5_history.json
Iteration 2¶
In [39]:
model,f=active_learning_iteration('resnet101_model_iteration_1.h5',model)
print()
print()
move_images_to_file(f)
print()
print()
model=train_model('resnet101_model_iteration_1.h5',model)
model_name = 'resnet101_model_iteration_2.h5'
model_path = os.path.join(models_folder, model_name)
model.save(model_path)
print(f'Model saved as {model_path}')
history_dict = history.history
history_filename = f'{model_name}_history.json'
with open(history_filename, 'w') as f:
json.dump(history_dict, f)
print(f'Training history saved as {history_filename}')
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Active Learning Iteration with model: resnet101_model_iteration_1.h5
C:\Users\mmac259\AppData\Roaming\Python\Python310\site-packages\keras\src\trainers\data_adapters\py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored. self._warn_if_super_not_called()
7/7 ━━━━━━━━━━━━━━━━━━━━ 11s 1s/step ['2174.png', '2294.png', '2525.png', 'Augmented_126_3213986.png', '2366.png', 'Augmented_132_16501.png', '2283.png', '2461.png', '2499.png', '2313.png', 'Augmented_124_438107.png', 'Augmented_161_612594.png', '2306.png', 'Augmented_160_760182.png', '2460.png', '2156.png', '2479.png', 'Augmented_202_319856.png', '2472.png', '2348.png', 'Augmented_124_9505309.png', 'Augmented_155_2219348.png', '2341.png', 'Augmented_142_9098843.png', '2354.png', '2074.png', 'Augmented_126_6286178.png', '2516.png', '2387.png', 'Augmented_131_8577086.png', '2518.png', '2400.png', 'Augmented_202_9754097.png', 'Augmented_152_9503415.png', '2512.png', '2295.png', '2381.png', '2337.png', '2336.png', '2517.png', '2361.png', '2316.png', 'Augmented_125_9536470.png', 'Augmented_122_2030302.png', 'Augmented_160_9074928.png', 'Augmented_127_4223591.png', 'Augmented_130_4523111.png', 'Augmented_168_636700.png', 'Augmented_125_9871364.png', 'Augmented_196_5628361.png', 'Augmented_138_321807.png', '2368.png', '2528.png', 'Augmented_124_5946084.png', 'Augmented_143_5910749.png', '2340.png', 'Augmented_201_8404995.png', '2393.png', 'Augmented_139_8251949.png', 'Augmented_189_1223578.png', 'Augmented_851_4702727.png', 'Augmented_158_3170394.png', 'Augmented_162_3161124.png', 'Augmented_125_5523914.png', 'Augmented_127_5342643.png', 'Augmented_132_9142230.png', '2489.png', 'Augmented_126_782575.png', '2334.png', '2383.png', 'Augmented_186_3518937.png', '2511.png', '2139.png', '2099.png', 'Augmented_121_7805233.png', '2377.png', 'Augmented_135_4830087.png', 'Augmented_159_4477058.png', '2510.png', '2386.png', 'Augmented_156_8313066.png', '2107.png', '2502.png', 'Augmented_160_3923740.png', '2352.png', 'Augmented_154_653776.png', '2308.png', 'Augmented_128_6121313.png', '2112.png', 'Augmented_158_1152895.png', '2467.png', '2485.png', 'Augmented_163_4743663.png', '2465.png', 'Augmented_123_8943539.png', 'Augmented_131_3782235.png', '2068.png', '2122.png', 'Augmented_166_5730205.png', '2346.png'] Selected files have been successfully moved to their respective directories.
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Found 3099 images belonging to 2 classes. Epoch 1/5 97/97 ━━━━━━━━━━━━━━━━━━━━ 103s 918ms/step - accuracy: 0.8477 - loss: 0.3539 - val_accuracy: 0.8297 - val_loss: 0.3704 Epoch 2/5 97/97 ━━━━━━━━━━━━━━━━━━━━ 88s 882ms/step - accuracy: 0.8761 - loss: 0.3014 - val_accuracy: 0.8250 - val_loss: 0.3707 Epoch 3/5 97/97 ━━━━━━━━━━━━━━━━━━━━ 88s 885ms/step - accuracy: 0.8763 - loss: 0.3116 - val_accuracy: 0.8422 - val_loss: 0.3527 Epoch 4/5 97/97 ━━━━━━━━━━━━━━━━━━━━ 87s 869ms/step - accuracy: 0.8633 - loss: 0.3248 - val_accuracy: 0.8406 - val_loss: 0.3437 Epoch 5/5 97/97 ━━━━━━━━━━━━━━━━━━━━ 87s 873ms/step - accuracy: 0.8716 - loss: 0.3193 - val_accuracy: 0.8406 - val_loss: 0.3456
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
Model saved as ./resnet101_model_iteration_2.h5 Training history saved as resnet101_model_iteration_2.h5_history.json
Iteration¶
In [41]:
model,f=active_learning_iteration('resnet101_model_iteration_2.h5',model)
print()
print()
move_images_to_file(f)
print()
print()
model=train_model('resnet101_model_iteration_2.h5',model)
model_name = 'resnet101_model_iteration_3.h5'
model_path = os.path.join(models_folder, model_name)
model.save(model_path)
print(f'Model saved as {model_path}')
history_dict = history.history
history_filename = f'{model_name}_history.json'
with open(history_filename, 'w') as f:
json.dump(history_dict, f)
print(f'Training history saved as {history_filename}')
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Active Learning Iteration with model: resnet101_model_iteration_2.h5 WARNING:tensorflow:5 out of the last 18 calls to <function TensorFlowTrainer.make_predict_function.<locals>.one_step_on_data_distributed at 0x0000021A74AD2B90> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
WARNING:tensorflow:5 out of the last 18 calls to <function TensorFlowTrainer.make_predict_function.<locals>.one_step_on_data_distributed at 0x0000021A74AD2B90> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
4/4 ━━━━━━━━━━━━━━━━━━━━ 9s 2s/step ['Augmented_157_3571532.png', '2307.png', '2519.png', 'Augmented_189_8372816.png', '2351.png', 'Augmented_165_8927752.png', '2152.png', 'Augmented_135_8205428.png', '2343.png', '2344.png', '2320.png', 'Augmented_122_32642.png', 'Augmented_130_8130609.png', 'Augmented_123_234492.png', 'Augmented_193_2525306.png', 'Augmented_124_677991.png', 'Augmented_190_4756924.png', 'Augmented_200_4695730.png', 'Augmented_162_2201276.png', 'Augmented_131_9731382.png', '2104.png', '2113.png', 'Augmented_127_9987249.png', 'Augmented_126_9710655.png', '2526.png', '2288.png', '2360.png', '2373.png', '2338.png', '2329.png', 'Augmented_155_932360.png', 'Augmented_191_1202396.png', 'Augmented_163_2756592.png', '2300.png', 'Augmented_156_2961688.png', '2370.png', 'Augmented_166_680824.png', '2384.png', 'Augmented_203_418066.png', '2363.png', 'Augmented_168_5501285.png', '2356.png', 'Augmented_197_5682271.png', 'Augmented_145_7840110.png', 'Augmented_161_199132.png', '2153.png', 'Augmented_121_7921024.png', '2136.png', 'Augmented_126_6879899.png', 'Augmented_200_4548100.png', '2515.png', '2374.png', '2380.png', 'Augmented_161_6597288.png', '2114.png', 'Augmented_190_3894107.png', 'Augmented_160_5572419.png', '2297.png', '2405.png', '2372.png', 'Augmented_126_9189439.png', '2476.png', 'Augmented_156_5051636.png', '2365.png', '2498.png', '2347.png', '2359.png', 'Augmented_188_8717618.png', '2529.png', '2345.png', '2396.png', '2395.png', 'Augmented_157_5707542.png', '2304.png', 'Augmented_202_6598051.png', 'Augmented_124_3480270.png', '2382.png', 'Augmented_123_4334518.png', '2111.png', '2491.png', 'Augmented_128_3785325.png', 'Augmented_122_1658862.png', 'Augmented_129_1486749.png', 'Augmented_205_830990.png', '2463.png', '2299.png', '2353.png', 'Augmented_125_2396264.png', 'Augmented_129_7810432.png', '2379.png', '2357.png', '2364.png', 'Augmented_134_9194421.png', '2076.png', 'Augmented_851_8447909.png', '2311.png', '2385.png', 'Augmented_155_3575357.png', '2376.png', '2258.png'] Selected files have been successfully moved to their respective directories.
WARNING:absl:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.
Found 3199 images belonging to 2 classes. Epoch 1/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 102s 883ms/step - accuracy: 0.8730 - loss: 0.3025 - val_accuracy: 0.8453 - val_loss: 0.3366 Epoch 2/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 88s 845ms/step - accuracy: 0.8794 - loss: 0.2836 - val_accuracy: 0.8281 - val_loss: 0.3571 Epoch 3/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 89s 853ms/step - accuracy: 0.8842 - loss: 0.2679 - val_accuracy: 0.8188 - val_loss: 0.3616 Epoch 4/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 90s 868ms/step - accuracy: 0.8650 - loss: 0.3134 - val_accuracy: 0.8109 - val_loss: 0.4318 Epoch 5/5 100/100 ━━━━━━━━━━━━━━━━━━━━ 116s 1s/step - accuracy: 0.8732 - loss: 0.2922 - val_accuracy: 0.8250 - val_loss: 0.3866
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.
Model saved as ./resnet101_model_iteration_3.h5 Training history saved as resnet101_model_iteration_3.h5_history.json
In [42]:
test_loss, test_acc = model.evaluate(test_gen, verbose=2)
print(f'Final Model - Loss: {test_loss}, Accuracy: {test_acc:.4f}')
results = {
'test_loss': test_loss,
'test_accuracy': test_acc
}
results_filename = f'{model_name}_final_results.json'
with open(results_filename, 'w') as f:
json.dump(results, f)
print(f'Test results saved as {results_filename}')
20/20 - 17s - 848ms/step - accuracy: 0.8234 - loss: 0.4129 Final Model - Loss: 0.4128965437412262, Accuracy: 0.8234 Test results saved as resnet101_model_iteration_3.h5_final_results.json
In [45]:
import os
from tensorflow.keras.preprocessing.image import load_img, img_to_array
images_folder = 'NewTest'
IMG_HEIGHT = 128
IMG_WIDTH = 128
def load_and_preprocess_image(image_path):
img = load_img(image_path, target_size=(128, 128))
img_array = img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = img_array / 255.0
return img_array
image_paths = [os.path.join(images_folder, img_name) for img_name in os.listdir(images_folder)]
with_mask_images = []
without_mask_images = []
for image_path in image_paths:
try:
img_array = load_and_preprocess_image(image_path)
prediction = model.predict(img_array)
label = 'Without Mask' if prediction > 0.5 else 'With Mask'
print(f"Image: {image_path}, Prediction: {prediction}, Label: {label}")
if label == 'Without Mask':
without_mask_images.append(img_array[0])
else:
with_mask_images.append(img_array[0])
except Exception as e:
print(f"Error loading image {image_path}: {e}")
def plot_images(images_arr, labels):
fig, axes = plt.subplots(1, len(images_arr), figsize=(20, 20))
axes = axes.flatten()
for img, ax, label in zip(images_arr, axes, labels):
ax.imshow(img)
ax.axis('off')
ax.set_title(label)
plt.tight_layout()
plt.show()
if with_mask_images:
plot_images(with_mask_images, ['With Mask'] * len(with_mask_images))
if without_mask_images:
plot_images(without_mask_images, ['Without Mask'] * len(without_mask_images))
WARNING:tensorflow:5 out of the last 12 calls to <function TensorFlowTrainer.make_predict_function.<locals>.one_step_on_data_distributed at 0x0000021AD9BF53F0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
WARNING:tensorflow:5 out of the last 12 calls to <function TensorFlowTrainer.make_predict_function.<locals>.one_step_on_data_distributed at 0x0000021AD9BF53F0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
1/1 ━━━━━━━━━━━━━━━━━━━━ 3s 3s/step Image: NewTest\178.png, Prediction: [[0.0108748]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Image: NewTest\187.png, Prediction: [[0.08122314]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 91ms/step Image: NewTest\190.png, Prediction: [[0.08247141]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 96ms/step Image: NewTest\219.png, Prediction: [[0.14663976]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Image: NewTest\231.png, Prediction: [[0.07945562]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 104ms/step Image: NewTest\283.png, Prediction: [[0.9885069]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 89ms/step Image: NewTest\350.png, Prediction: [[0.00109553]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 108ms/step Image: NewTest\351.png, Prediction: [[0.4990173]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 85ms/step Image: NewTest\69.png, Prediction: [[0.952248]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 96ms/step Image: NewTest\72.png, Prediction: [[0.9874674]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Image: NewTest\download (1).jpg, Prediction: [[0.23219408]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 93ms/step Image: NewTest\download (2).jpg, Prediction: [[0.1215727]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 87ms/step Image: NewTest\download (3).jpg, Prediction: [[0.5045084]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 97ms/step Image: NewTest\download (4).jpg, Prediction: [[0.04255181]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 88ms/step Image: NewTest\download (6).jpg, Prediction: [[0.01247125]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 90ms/step Image: NewTest\download (7).jpg, Prediction: [[0.82474047]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Image: NewTest\download.jpg, Prediction: [[0.06365281]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 100ms/step Image: NewTest\face-13.jpg, Prediction: [[0.9260524]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 84ms/step Image: NewTest\images (2).jpg, Prediction: [[0.01780668]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 91ms/step Image: NewTest\images (3).jpg, Prediction: [[0.2885575]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 94ms/step Image: NewTest\images (4).jpg, Prediction: [[0.04444886]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Image: NewTest\images.jpg, Prediction: [[0.6170349]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Image: NewTest\jh.jpg, Prediction: [[0.12678163]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 83ms/step Image: NewTest\kjk.jpg, Prediction: [[0.02638604]], Label: With Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step Image: NewTest\lk.jpg, Prediction: [[0.05673988]], Label: With Mask Error loading image NewTest\Thumbs.db: cannot identify image file <_io.BytesIO object at 0x0000021AD984DD50> 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 86ms/step Image: NewTest\uu.jpg, Prediction: [[0.5006125]], Label: Without Mask 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 89ms/step Image: NewTest\with.jpg, Prediction: [[0.00311736]], Label: With Mask
